Legal AI Vendor Scores

Not found in public documents

No official Aira page names the AI model or provider behind the receptionist; the site only says it uses the latest AI technology. The Terms also say Aira can change the Terms without prior notice, so there is no commitment to tell customers before material changes.

• “Natural, human-like voices powered by the latest AI technology.” Aira AI Receptionist
• “Aira reserves the right, at its sole discretion, to modify these Terms at any time and without prior notice.” Aira Terms of Service

Ask the vendor: Which AI model providers (for language, speech, and voice) power Aira, and will you commit in writing to notify us before changing them or making material changes to the service?

A firm can sign up for a monthly plan starting at $24.95 with no long-term contract and cancel anytime, and the site also offers a self-use live voice demo. No free trial is described on official pages.

• “No long-term contracts required” Aira Brand Facts
• “Compare four Aira plans from $24.95/month. ... Cancel anytime.” AI Receptionist Pricing
• “Live browser conversation · No scripted video · Test the experience yourself” Aira product demo

Addressed, but falls short of the standard

This is still a Fail under the amended rule. Aira does not say it only uses anonymized or aggregated data. Its Terms give Aira a permanent license to use customer content itself, including call recordings, to improve the service. No official source says call data is kept out of AI training by Aira or by its AI providers. The only no-training promise in the Privacy Policy covers Google Workspace data.

• “perpetual, irrevocable license to use, store, reproduce, modify, create derivative works from ... such User Content as necessary to operate and improve the Service.” Aira Terms of Service
• “These recordings are provided for your convenience and to improve the Service” Aira Terms of Service

Ask the vendor: Will you confirm in writing that our call recordings, transcripts, and intake data are never used in identifiable form to train or improve Aira's models or any third-party AI provider's models, and that your AI providers are contractually bound to no training?

Addressed, but falls short of the standard

The Privacy Policy gives only an open-ended retention standard. No deletion timeline, backup deletion schedule, or proof of destruction is published.

• “We retain information for as long as necessary to provide the Service, fulfill the purposes outlined in this Privacy Policy, comply with legal obligations” Aira Privacy Policy

Ask the vendor: How many days after we request deletion or close our account are our recordings, transcripts, and backups permanently deleted, and will you provide written confirmation?

Not found in public documents

This is still a Fail. The new benefit of the doubt applies only when a vendor claims SOC 2, and Aira makes no SOC 2 claim anywhere on its site. Its security, trust, DPA and SOC 2 pages return not found. No DPA is published or offered on request. Aira's own HIPAA page says it does not sign BAAs, which matters for personal injury firms whose callers describe injuries and medical treatment.

• “As of April 2026, Aira is not HIPAA covered and does not sign Business Associate Agreements with healthcare providers.” Aira HIPAA page

Ask the vendor: Do you have a current SOC 2 Type II report covering the AI receptionist, and can you provide a DPA (and a BAA, if callers share health information) before we sign up?

Not found in public documents

Aira lists only general categories of vendors, not a named subprocessor list. It makes no commitment to give notice before adding one and names no data location. It says data may be processed in other countries.

• “With third-party vendors or contractors who assist with payment processing, data analytics, hosting, customer support, marketing, or other services” Aira Privacy Policy
• “Your information may be transferred to and processed in countries other than your own.” Aira Privacy Policy

Ask the vendor: Can you send your list of named subprocessors (including AI and telephony providers), tell us where our data is stored, and commit to notifying us before adding new ones?

Not found in public documents

This is still a Fail. The law firm page mentions encryption and dashboard access, but nothing says one customer's data is kept separate from another's. Aira also has no SOC 2 claim, pen test or red-team exercise that could count as independent support.

• “All call data is encrypted, stored securely, and accessible only through your password-protected dashboard.” AI Receptionist for Law Firms

Ask the vendor: How is our firm's data separated from other customers' data, and what current independent audit or penetration test confirms that separation?

Not found in public documents

Neither the Terms nor the Privacy Policy contains any commitment to notify customers of a data breach, and no timeframe is stated. The only published contact is hey@getaira.io.

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a breach affecting our data, and who is the contact?

Not found in public documents

No official source describes exporting call records, recordings, or transcripts in a usable format, and none promises deletion afterward. The Terms warn that termination may cut off access to your content.

• “If your account is terminated, you may lose access to User Content and other information associated with your account.” Aira Terms of Service

Ask the vendor: Can we export all recordings, transcripts, and intake records in a standard format before leaving, and will you confirm deletion in writing afterward?

Nothing public describes when Aira staff can access customer call data, whether that access is logged, or whether the firm's permission is required. The team accounts page covers only the firm's own team members.

Ask the vendor: Which Aira employees can access our call recordings and transcripts, is that access logged, and do you need our permission first?

Every call is recorded and transcribed and can be searched in the dashboard, which covers what the AI said to callers. No official source confirms a customer-exportable log.

• “Every call is recorded with a full transcript. Listen back, search conversations, and train your team.” Aira Features

Ask the vendor: Can we export a complete log of every call transcript and AI response, with dates and the user or line it is tied to?

Aira talks with callers live and sends texts and follow-up sequences on its own. The firm approves message templates and stop conditions ahead of time, but individual messages are not reviewed before they reach callers. No account-wide off switch is described.

• “Review the message, timing, and the moments when Aira should stop and hand the conversation back to a person” AI receptionist automations
• “Your automation stops when the customer or your team replies, an appointment is booked, or the contact opts out” AI receptionist automations

Ask the vendor: What can Aira say or send to prospective clients without a person reviewing it, and how do we immediately turn off all AI calls and texts if needed?

Not found in public documents

Alexi never names the AI model or model company behind the product. Its FAQ only says it uses outside models that do not keep data. The DPA lists Azure and AWS for 'Model Hosting' but does not say which model runs there. The Terms do promise 30 days' notice before changes to the agreement. The change notice is there, but the model name is missing, so this fails.

• “Alexi uses zero-data-retention models (no prompts, documents, or outputs are retained by the model provider).” Alexi FAQs
• “Alexi may unilaterally amend this Agreement... by giving Customer 30 days prior notice of such Amendment or posting notice” Alexi Terms of Service

Ask the vendor: Which company's large language model does Alexi use today (for example OpenAI through Azure, or Anthropic), and will you tell us in writing before you switch?

Alexi advertises a guided trial with hands-on access after a short walkthrough call, and its Terms include a trial clause. The trial page talks about seeing value 'in just a few days,' and the FAQ says evaluations usually take about 30 days. No exact trial length is published, and the trial is set up through an Order Form. The Terms warn that data entered during a trial is lost unless you subscribe, and that Alexi can end a trial at any time.

• “Get hands-on access to Alexi with a quick guided walkthrough. Ask real legal questions, explore your workflows” Alexi Start Your Guided Trial page
• “If a Trial Period is specified in an applicable Order Form, Alexi may make the applicable Alexi Solution available to Customer on a trial basis” Alexi Terms of Service, Section 2(g) Trials

Vendor's own documents contradict each other

The new anonymized-data allowance does not rescue Alexi. The model provider half is covered: Alexi says it uses zero-data-retention models. But the Terms give Alexi a permanent license to use your inputs and data themselves to improve Alexi's products and other offerings. Only a separate part of that clause, about usage data, requires anonymizing. The Terms also define AI models trained on Customer Data and say Alexi owns them. That conflicts with the FAQ promise that your data never trains global models.

• “Your prompts, documents, and workflow data never train global models or mix with other firms.” Alexi FAQs
• “perpetual, royalty-free, irrevocable... license to... Process Customer Property to: (i) improve and enhance the Alexi Solution and Alexi's other offerings” Alexi Terms of Service, Section 3
• “"Custom GenAI Model" means any GenAI model that has been trained or fine-tuned using Customer Data.” Alexi Terms of Service, Definitions
• “Alexi uses zero-data-retention models (no prompts, documents, or outputs are retained by the model provider).” Alexi FAQs

Ask the vendor: Will you put in the signed contract that our identifiable inputs, outputs, and documents will not be used to train or improve any model, limiting the Section 3 improvement license to anonymized data and excluding 'Custom GenAI Models'?

Addressed, but falls short of the standard

Alexi says it deletes customer data when you leave. No published document gives a deletion deadline, a backup purge schedule, or a certificate of destruction. Backups may be kept until deleted 'in the ordinary course,' and the DPA allows retention for open-ended legal and dispute reasons.

• “Alexi may retain Customer Data in its backups, archives and disaster recovery systems until such Customer Data is deleted in the ordinary course” Alexi Terms of Service, Section 13(c)
• “purges or removes customer data containing confidential information from the application environment, in accordance with best practices, when customers leave” Alexi Trust Center Controls

Ask the vendor: Within how many days of termination or a deletion request will our data be deleted from live systems and from backups, and will you give us written confirmation of destruction?

Alexi's Trust Center lists a SOC 2 Type II report you can request. Under the amended rule, we assume it is current and covers the product, even though no audit date is published. The DPA is published on Alexi's website, and it mentions that a HIPAA business associate agreement can be signed with Alexi.

• “Alexi SOC 2 Type II Report.pdf ... Request access” Alexi Trust Center Resources
• “without first either (i) deidentifying such information... or (ii) entering into a "business associate agreement" with Alexi” Alexi Data Processing Addendum (2025)

Vendor's own documents contradict each other

All three pieces are published, but they disagree. DPA Section 4.2 promises notice before adding a subprocessor and gives you 30 days to object. The DPA's subprocessor list includes AWS and SerpAPI and places hosting in Canada. The Trust Center list leaves out AWS and SerpAPI and adds Weaviate, Vanta, and Heroku (Heroku in Canada, the EU, and the US). Yet the Trust Center FAQ says Alexi's servers are on AWS. Neither list names an AI model provider.

• “AWS Cloud Infrastructure Services and Model Hosting Canada” Alexi Data Processing Addendum, Annex B and Section 4.2
• “Microsoft Azure, Weviate, Vanta, GitHub, Bitbucket, Heroku (1-6 of 6 results)” Alexi Trust Center Subprocessors
• “Alexi's servers are located in AWS, with options for data residency in the CA, EU, or US regions.” Alexi Trust Center FAQ

Ask the vendor: Which subprocessor list is current and binding (the DPA Annex B or the Trust Center), which AI model provider handles our data, and in which country will our data be stored and processed?

Alexi states that each firm's data is kept separate, with no access between firms. The Trust Center lists a SOC 2 Type II report and a 2025 penetration test. Under the amended rule, the SOC 2 Type II claim counts as current independent support. Firms that want more can choose a fully isolated private cloud deployment.

• “Strict data separation per firm, no cross-tenant access, inference-only model usage.” Alexi Security page
• “Pen Test 2025 ... Alexi SOC 2 Type II Report.pdf ... Request access” Alexi Trust Center Resources

Addressed, but falls short of the standard

The DPA promises written notice of a breach 'without undue delay' but gives no number of hours. The Terms and Privacy Policy give no breach timing either. The Trust Center lists security@alexi.com as a contact and an Incident Response Plan you can request.

• “Alexi shall notify Customer in writing without undue delay upon Alexi becoming aware of a Privacy Breach.” Alexi Data Processing Addendum, Section 5

Ask the vendor: Will you commit in the contract to notifying us of any breach involving our data within 72 hours, and who will contact us?

Addressed, but falls short of the standard

You can retrieve your data for 30 days after termination, but only if you ask in writing at least 30 days before the end and have paid all fees. No file format is named. The DPA promises disposal, with exceptions, and backups may be kept until deleted 'in the ordinary course.' Nothing mentions proof of destruction.

• “Alexi will make all Customer Data available to Customer for electronic retrieval for a period of 30 days.” Alexi Terms of Service, Section 13(c)
• “Alexi will dispose of (or, at Customer's written request, return) the Personal Information, subject to Alexi's requirements to retain” Alexi Data Processing Addendum, Section 6

Ask the vendor: In what file formats can we export our documents, prompts, and outputs, and will you give us written certification that all copies, including backups, are destroyed afterward?

Staff access is limited to people with a business need who are bound by confidentiality. Nothing published says staff need your firm's permission, that access to your data is logged for you, or that you can revoke it.

• “authorize access to Personal Information by its employees and agents only if (i) they need to have access” Alexi Data Processing Addendum, Section 2.3(h)
• “The company restricts privileged access to databases to authorized users with a business need.” Alexi Trust Center Controls

Ask the vendor: Do Alexi staff need our explicit permission before viewing our data, is that access logged, and can we revoke it?

Firm admins can view prompts and workflow output logs in a central console. Nothing published confirms that the log covers every prompt and answer, ties each to a user, or can be exported.

• “Admins can view prompts, usage metadata, workflow output logs, and access settings in a centralized console.” Alexi FAQs

Ask the vendor: Can we export a complete log of every prompt and AI answer, tied to the user who made it?

Alexi offers agents and email automation. It mentions approvals inside workflows and lets firms disable features or limit workflows, but only in general terms. Nothing published lists what the AI can do without a person approving it, or explains how to stop an automation that is running.

• “turns playbooks, templates, and approvals into intelligent workflows that scale expertise while maintaining complete control” Alexi Agents & Workflow Automations page
• “Firms can disable features, restrict data sources, or limit workflows based on governance needs.” Alexi FAQs

Ask the vendor: Which Alexi agents or email automations can act without a lawyer approving first, and how do we stop one immediately?

Not found in public documents

No official Briefpoint page names the AI model or provider behind the product. The Terms, Privacy Policy, FAQ and Trust Center subprocessor list are all silent on this, and the subprocessor list names no AI provider at all. The Terms do promise notice before changes to the agreement. Without the model named, this fails.

• “Briefpoint may modify this Agreement from time to time with notice to Customer.” Briefpoint Terms & Conditions

Ask the vendor: Which company's large language model powers Briefpoint (for example OpenAI, Azure OpenAI, or Anthropic), and will you tell us in writing before you switch it?

Not found in public documents

The only published plan is billed annually, and the website offers only "Book a Demo". The Terms describe how trials work (30 days by default) if one is granted, but no official page says a prospective customer can actually get a trial or other hands-on access before signing an annual contract.

• “$150/month per attorney when billed annually” Briefpoint Pricing
• “use is permitted only for Customer's internal evaluation during the period designated by Briefpoint (or if not designated, 30 days).” Briefpoint Terms & Conditions

Ask the vendor: Can our firm get hands-on access to the real product, such as a 30-day trial, before signing an NDA or an annual contract?

The FAQ says nothing you type is used to train Briefpoint's own models or any third-party model, which covers both halves. The Terms limit what Briefpoint can use to improve its product to usage data, which expressly excludes your inputs and outputs. Worth knowing: the no-training promise sits in the FAQ rather than the contract, and the AI provider is not named.

• “Nothing you type into Briefpoint is used to train Briefpoint or any third-party AI model.” Briefpoint FAQ
• “Never used for machine learning or model training” Briefpoint FAQ
• “learnings about Users' use of the Service, but excluding any Input or Output.” Briefpoint Terms & Conditions

Addressed, but falls short of the standard

The Terms name a 7-day deletion window after cancellation, but the word "Generally" leaves an undefined exception, and the Terms let copies stay in backups even after deletion with no schedule for purging them. No certificate of deletion is offered. The FAQ says deleted data is permanently removed but gives no timeframe.

• “Generally, Customer's data will be automatically deleted within 7 days of cancellation of services.” Briefpoint Terms & Conditions
• “may be retained in Recipient's standard backups notwithstanding any obligation to delete” Briefpoint Terms & Conditions
• “We retain your personal information for as long as necessary to provide the Service” Briefpoint Privacy Policy

Ask the vendor: When are exceptions made to the 7-day post-cancellation deletion, on what schedule are backups purged, and will you confirm deletion in writing?

Briefpoint claims SOC 2 Type II and posts an attestation letter, which under the amended rule is treated as current. The FAQ also says a BAA is available on request, so the health-information piece is covered, although the Terms say Briefpoint is not a HIPAA Business Associate. The remaining gap is the Data Processing Agreement: none is published on the website, in the Terms or Privacy Policy, and none is listed among the Trust Center's requestable documents (checked 2026-10-06). Rule update: a DPA is no longer required to pass. If a vendor has one, it should make it available on request, so a firm should ask for it before signing.

• “SOC2 Type II Letter of Attestation” Briefpoint Trust Center, Resources
• “Contact your account manager or our support team to request a BAA.” Briefpoint FAQ
• “Briefpoint is not a Business Associate as defined under HIPAA.” Briefpoint Terms & Conditions

Addressed, but falls short of the standard

The Trust Center lists five subprocessors, but none is an AI model provider and Amazon is missing, even though the Privacy Policy says data is stored mostly on Microsoft and Amazon servers. No document commits to notifying customers before a new subprocessor is added, and no data location (country or region) is stated.

• “Microsoft Azure, Cloud provider; Google Workspace; Vanta; GitHub; Auth0” Briefpoint Trust Center, Subprocessors
• “predominantly servers owned by Microsoft and Amazon” Briefpoint Privacy Policy

Ask the vendor: Which AI provider and hosting regions process our data, why are they missing from your subprocessor list, and will you notify us before adding new subprocessors?

The FAQ says each account's data is kept separate. Briefpoint claims SOC 2 Type II and posts an attestation letter from Prescient Assurance. Under the amended rule this counts as current independent support, even though the posted audit period ended in September 2025.

• “Client data is siloed per account and never shared” Briefpoint FAQ
• “operated effectively throughout the period May 20, 2025 to September 3, 2025” SOC2 Type II Letter of Attestation (Prescient Assurance)

Not found in public documents

Neither the Terms, Privacy Policy, FAQ nor Trust Center gives any time frame for notifying customers of a data breach. The Terms require the customer to notify Briefpoint of compromised logins, but contain no matching duty on Briefpoint. The only published contact is support@briefpoint.ai.

Ask the vendor: Will you commit in writing to notify us of any security breach involving our data within 72 hours, and who would contact us?

Addressed, but falls short of the standard

The Terms tell customers to download their data before cancelling and say data is generally deleted within 7 days afterward. The FAQ says documents export as Word files. No document says what format a full account export comes in, the deletion promise is hedged with "Generally" and a backup exception, and no proof of destruction is offered.

• “It is recommended that Customers download any data they need to retain post-termination before submitting a cancellation.” Briefpoint Terms & Conditions

Ask the vendor: Can we export all our matters and documents in standard formats at the end of the contract, and will you confirm in writing that all copies, including backups, were destroyed?

The Privacy Policy limits staff access to "Engineering Reviews", which customers can opt out of by emailing support, and the Trust Center lists access restriction controls. Access is on by default rather than requiring your permission each time, and nothing says staff access is logged.

• “Client Data submitted to Briefpoint is not accessible by the Briefpoint admins outside of Engineering Reviews” Briefpoint Privacy Policy
• “The company restricts privileged access to databases to authorized users with a business need.” Briefpoint Trust Center, Controls

Ask the vendor: Is staff access to our client data logged, and can we require our explicit permission before each Engineering Review?

Nothing published describes a log of AI prompts and answers, or any way for customers to export one.

Ask the vendor: Can we export a log of every AI request and response, showing which user made it?

Discovery drafts come out as Word documents for the attorney to review before serving, and the Terms put responsibility for checking output on the customer. The client collection feature sends clients a secure link to answer interrogatories, and nothing published says whether an attorney must approve those messages before they go out.

• “You review and revise in Word.” Briefpoint homepage
• “Briefpoint sends your client a secure link to answer interrogatories and upload documents” Briefpoint homepage
• “It is Customer's sole responsibility to verify the accuracy of any information obtained from the Service” Briefpoint Terms & Conditions

Ask the vendor: Does an attorney have to review and approve the questions and links sent to our clients through client collection before they go out?

Addressed, but falls short of the standard

StrongSuit names its model providers (OpenAI, Anthropic, Google, and others) in the Terms of Service and Privacy Policy. But neither document promises to tell customers before material changes. The Terms take effect the moment they are posted, and the Privacy Policy mostly notifies by posting changes on its own page, with email only 'in some cases'.

• “modules routing through external providers (e.g., OpenAI, Anthropic, and Google)” StrongSuit Terms of Service
• “StrongSuit may change these Terms by posting updated Terms on its Website. All updates become effective when posted.” StrongSuit Terms of Service

Ask the vendor: Will you notify us in advance, by email, before you change the AI model provider or make material changes to the service or these terms?

You can sign up yourself for a free trial, and the Individual plan is billed monthly at $249 with cancel anytime. A credit card is required, and the trial turns into a paid plan automatically if you don't cancel. The site navigation refers to a 5-day free trial, but the Terms and pricing page do not state the length.

• “A valid credit card is required to begin the free trial” StrongSuit Terms of Service
• “Cancel anytime” StrongSuit Pricing Plans

The Privacy Policy says client data is not used to train StrongSuit's own models or any third-party models, including those of its sub-processors. That covers both halves. The Terms also let StrongSuit share technical usage data that is aggregated and anonymized, which is acceptable under the amended rule. The policy does not say whether the AI providers keep data after processing it.

• “We do not and will not use Client Data to train, retrain, or improve any proprietary or third-party large language models” StrongSuit Data Privacy Policy
• “including those provided by sub-processors” StrongSuit Data Privacy Policy
• “StrongSuit will not train any artificial intelligence (AI) models with User Content.” StrongSuit Terms of Service

Addressed, but falls short of the standard

For client data, both documents reserve a right to delete it after the subscription ends. Neither commits to deleting it by a set date, and nothing covers backups or proof of deletion. Account and identity data has a firmer promise: it is deleted within 90 days.

• “permanently delete all Client Data remaining in our possession at any time after the subscription ends” StrongSuit Data Privacy Policy
• “StrongSuit may securely delete all remaining User Content and will have no obligation to retain or provide access” StrongSuit Terms of Service

Ask the vendor: Within how many days of termination or a deletion request will you delete our client data, including backups, and will you confirm the deletion in writing?

The pricing page claims SOC 2 Type II, which now counts as current under the benefit of the doubt. The fail is the DPA. No DPA is published, none is listed as requestable, and neither the Terms, the Privacy Policy, nor the security page mentions one (rechecked 2026-10-06). The security page offers only a general contact address. StrongSuit is a research and drafting tool, not a dedicated health-records product, so the missing BAA alone would not cause a Fail. Firms handling PI or medical files should still ask about one. Rule update: a DPA is no longer required to pass. If a vendor has one, it should make it available on request, so a firm should ask for it before signing.

• “Models never trained on user data + SOC 2 Type II” StrongSuit Pricing Plans
• “StrongSuit is SOC2 certified, offering independent third-party verification of our operational and security controls.” Security & Confidentiality at StrongSuit

Addressed, but falls short of the standard

Data location is published (United States only). The Privacy Policy names Microsoft Azure, OpenAI, Google, and Anthropic 'and others', so the list is incomplete. The Terms say StrongSuit may use subcontractors as it deems appropriate, and nothing promises notice before new subprocessors are added.

• “All Client Data is stored on secure, geographically redundant servers located exclusively in the United States.” StrongSuit Data Privacy Policy
• “third-party LLMs provided by OpenAI, Google, Anthropic, and others” StrongSuit Data Privacy Policy

Ask the vendor: Can you give us your complete subprocessor list and commit to notifying us before you add a new one?

Addressed, but falls short of the standard

The claimed SOC 2 Type II now counts as independent support, so the remaining gap is the isolation statement itself. The security page talks about preventing leakage 'between sessions'. The private-tenant language describes keeping data away from the AI providers, not separating one customer from another. A 2026-10-06 recheck of the security page and Privacy Policy found no statement that one firm's data is separated from other customers' data.

• “session management and data-level encryption to prevent leakage between sessions” Security & Confidentiality at StrongSuit
• “Data processing for these models occurs exclusively within a secure, private tenant environment” StrongSuit Data Privacy Policy

Ask the vendor: Can you confirm in writing that our firm's data is logically isolated from other customers' data, and that your SOC 2 Type II report covers that control?

Not found in public documents

None of the Terms of Service, the Privacy Policy, or the security page gives a timeframe for notifying customers of a data breach. The security page lists security@strongsuit.com as a general security contact.

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a breach affecting our data, and who will contact us?

Addressed, but falls short of the standard

Exporting your data before you leave is your job, but no source explains how to export it or in what format. After termination, StrongSuit only reserves a right to delete the data. It does not promise deletion or proof of destruction.

• “The Customer is ultimately responsible for the ongoing management and timely export of its Client Data prior to subscription termination.” StrongSuit Data Privacy Policy

Ask the vendor: In what file formats can we export all our matters, documents, and AI work product, and will you certify deletion afterward?

Staff access is limited to stated purposes, such as support you ask for, and the security page mentions audit trails. Other listed reasons do not need your permission, and nothing says you can revoke access.

• “Employees do not monitor Client Data, except when absolutely necessary to: (a) provide technical support explicitly requested by you” StrongSuit Data Privacy Policy
• “Detailed audit trails tracking infrastructure and application-level activity” Security & Confidentiality at StrongSuit

Ask the vendor: Will your staff access our data only with our explicit permission, is every access logged, and can we revoke it?

The only mention is an internal security audit trail of infrastructure and application activity. Nothing public describes an AI activity log that customers can see, a log of every AI prompt and answer tied to each user, or a way to export one.

• “Detailed audit trails tracking infrastructure and application-level activity” Security & Confidentiality at StrongSuit

Ask the vendor: Can we see and export a log of every AI prompt and response, showing which user ran it and when?

The homepage describes 'lawyer in the loop' workflows, and the product produces drafts, memos, and redlines for the lawyer to review. No published document describes formal approval gates, and none says whether any feature acts on its own or how to stop it.

• “every draft, review, and decision stays under your strategic control” StrongSuit homepage

Ask the vendor: Does any StrongSuit feature send, file, or share anything without a lawyer approving it first?

Not found in public documents

No official CaseGen page names the underlying language model or AI provider. The Subscription Agreement only refers to unnamed third-party AI platforms. The Terms promise reasonable efforts at 30 days' notice of material term changes, but the Subscription Agreement says changes can take effect on posting, and without a named model the first half of this question is not met.

• “MAF AI, INC may use third-party AI components or services in providing the Services.” Subscription Agreement

Ask the vendor: Which AI model provider (for example OpenAI, Anthropic, or Google) powers Justina, Justin, and Maya, and will you notify us in writing before you switch providers?

CaseGen publishes month-to-month plans with no long-term contract, so a firm can use the real product without an annual commitment. The Terms allow an optional free trial of up to 7 days, offered at the company's sole discretion. Note there is a one-time implementation fee, and the Subscription Agreement requires 30 days' written notice to cancel a month-to-month plan.

• “Plans are month to month, with no long-term contracts.” Pricing and implementation guide
• “offer a Subscription with a Free Trial for a limited period of time, not exceeding seven (7) days.” Terms of Service

Vendor's own documents contradict each other

Under the amended rule, CaseGen's use of anonymized or de-identified data (Section 15.1 and the Privacy Policy) would be acceptable. The problem is Section 7.1 of the same contract. It lets CaseGen use the firm's actual data and call recordings to train its own models and those of affiliated AI platforms, with no anonymization limit. A re-check of the agreement on 2026-10-06 confirmed this. That is a license to train on identifiable customer content, and it conflicts with Section 15.1 and the blog's 'never trains' promise, so it still fails.

• “MAF AI, INC may use Subscriber's data and call recordings to train, develop, and improve its machine learning models and algorithms” Subscription Agreement, Section 7.1
• “utilizing the data within MAF AI, INC's own machine learning systems and/or those of its affiliated artificial intelligence platforms” Subscription Agreement, Section 7.1
• “MAF AI, INC may use anonymized and aggregated data derived from Subscriber's use” Subscription Agreement, Section 15.1
• “CaseGen never trains on client data.” Vendor blog

Ask the vendor: Will you amend Section 7.1 in writing so that only anonymized or de-identified data, never our identifiable recordings, transcripts, or intake data, is used to train your models or any affiliated AI platform's models?

Addressed, but falls short of the standard

The Privacy Policy sets 90-day default retention for recordings and transcripts and says deletion requests are typically handled within 30 days. But the 30 days is described as typical, not committed, backups follow unstated practices, and there is no deletion certificate or verification step. The Subscription Agreement says data may be deleted at CaseGen's sole discretion after termination.

• “We will process deletion requests within a commercially reasonable timeframe (typically within 30 days)” Privacy Policy
• “will be deleted or overwritten in accordance with our backup retention practices.” Privacy Policy

Ask the vendor: Will you commit in writing to deleting our data, including backups, within a fixed number of days of our request and provide written confirmation once it is done?

Not found in public documents

The benefit of the doubt applies only when a vendor says it has SOC 2. CaseGen says only 'SOC 2 Ready' and 'compliant with SOC 2 standards.' That describes readiness, not a completed audit. CaseGen has no Trust Center (its /security/ page returns 404), publishes no DPA, and its contract and Privacy Policy never mention HIPAA or a BAA. The product collects medical treatment details for personal injury intake and coordinates medical appointments, so a BAA before signing is required, and none is confirmed. Even if the SOC 2 wording were read generously, the missing DPA and BAA would still mean a Fail.

• “HIPAA & SOC 2 Ready: Privacy and security baked in” Industries page
• “While it's not yet HIPAA-certified, it adheres to strict privacy standards” Vendor blog

Ask the vendor: Do you hold a completed SOC 2 Type II report, and can you send it along with your Data Processing Agreement and a signed Business Associate Agreement before we sign?

Addressed, but falls short of the standard

Data location (United States) is published and Google Cloud is named as the host. Other subprocessors are listed only by category (telephony, payments, analytics, support tools), the AI model provider is not named, and no document promises notice before new subprocessors are added.

• “CaseGen uses Google Cloud Platform (GCP) for hosting and data storage.” Privacy Policy
• “personal information and Customer Data processed through the Services are processed and stored in the United States” Privacy Policy

Ask the vendor: Can you provide a full named list of your subprocessors, including your AI model and telephony providers, and commit to notifying us before adding new ones?

Addressed, but falls short of the standard

The Privacy Policy says identifiable data is not shared across customers. Read generously, that counts as an isolation statement. But CaseGen claims only that it is 'SOC 2 Ready,' not that it has SOC 2, so the amended benefit of the doubt does not apply. CaseGen also publishes no penetration test or red-team exercise, so nothing independent supports the isolation statement.

• “CaseGen does not share identifiable Customer Data across customers or organizations.” Privacy Policy

Ask the vendor: How is our firm's data technically separated from other customers' data, and what independent audit or penetration test confirms it?

The Subscription Agreement commits to notifying the firm within 48 hours of becoming aware of a breach, with details the firm needs to report to authorities. The Privacy Policy separately says 72 hours where required by law; the contract's 48-hour term is the stronger, binding commitment. The contact channel listed in the Privacy Policy is support@casegen.ai.

• “notify Subscriber without undue delay, and in any event within 48 hours of becoming aware” Subscription Agreement, Section 14.5

Addressed, but falls short of the standard

The contract puts export on the firm but does not describe an export tool or file format, and deletion after termination is at CaseGen's discretion rather than a commitment to destroy data. A CaseGen blog article mentions nightly CSV exports for integrations, but no official contract or help page commits to customer export or proof of destruction.

• “Subscriber is solely responsible for exporting or retrieving any data prior to the effective date of termination.” Subscription Agreement, Section 3.4
• “may be deleted or destroyed at MAF AI, INC's sole discretion without further notice or liability” Subscription Agreement, Section 3.4

Ask the vendor: How can we export all our recordings, transcripts, and intake records in a usable format when we leave, and will you confirm in writing that our data is destroyed afterward?

The contract says access is limited to authorized personnel and the Privacy Policy mentions monitoring and logging. Nothing says staff need the firm's permission to access data or that the firm can revoke access. CaseGen also runs a managed service where its staff configure the system for the firm.

• “Access controls to limit access to Subscriber Data to authorized personnel” Subscription Agreement, Section 14.2

Ask the vendor: Do your staff need our explicit permission before viewing our call recordings or intake data, and can we see and revoke that access?

CaseGen describes real-time transcripts, call playback, automated case summaries, and call analytics. No official document confirms a complete, user-attributed log of every AI interaction that the firm can export.

• “Replay and review past calls anytime for training or case verification.” Homepage
• “Follow the transcript as the call is unfolding in real-time.” Homepage

Ask the vendor: Can we export a complete log of every AI call, transcript, and summary, tied to the caller and date, at any time?

The AI answers calls, makes outbound follow-up calls, and sends texts to prospective clients. The contract says outbound calls and texts happen only when the firm explicitly authorizes them, which is a general approval gate. CaseGen also states configurable stopping rules are not available and that firms oversee decisions to stop outreach, so no product-level stop mechanism is published.

• “Making outbound calls on behalf of Subscriber when explicitly authorized” Subscription Agreement, Section 7.3
• “configurable stopping rules are not currently available” llms.txt product summary

Ask the vendor: How do we pause or stop Justin's outbound calls and texts immediately, and can we require approval before any message goes to a client?

Not found in public documents

The Terms of Service give 30 days before material changes to the terms take effect, which satisfies the change-notice half. However, no official Clearbrief page, the Terms, or the Privacy Policy names the underlying AI model or provider. The homepage refers only to generative AI and large language models in general.

• “For any material changes to these Terms, such amended terms will automatically be effective thirty days after they are initially posted” Clearbrief Terms of Service, Section 4

Ask the vendor: Which company and model power Clearbrief's generative AI features, and will you notify us in writing before switching providers?

The Solo plan is offered at a monthly price with a self-serve purchase button, and the Terms allow monthly subscriptions to be cancelled with no cancellation charge. That is month-to-month hands-on access without an annual contract or NDA. No free trial is published on the vendor's own pages (the Terms mention 'any applicable free trial period' but no trial offer is stated). Fees already paid are non-refundable.

• “SOLO $300/month Per user per month PURCHASE NOW” Clearbrief homepage, Pricing
• “For monthly accounts or other accounts with a periodic subscription cycle of less than one year, there are no charges for cancelling a subscription.” Clearbrief Terms of Service, Section 8.6

Addressed, but falls short of the standard

The homepage says customer data is never used to train large language models, which covers Clearbrief's own side. No official source says the underlying AI provider is bound by a no-training or zero-retention agreement, and the provider is not even named, so the second half of the test is silent. Under the amended rule, the aggregate-use clause in Section 13.5 is acceptable. Section 13.3, however, licenses Clearbrief to use customer-provided information itself, not just de-identified or aggregated data, to create new features, which the amended rule still treats as a concern.

• “Your data is never used for training large language models.” Clearbrief homepage, Security section
• “perpetual, irrevocable, fully paid-up, and royalty-free license to use, your User Provided Information ... to create new features and functionality” Clearbrief Terms of Service, Section 13.3
• “to use such Information in aggregate form to improve the way the Clearbrief Services work and look” Clearbrief Terms of Service, Section 13.5

Ask the vendor: Who is your AI model provider, and will you confirm in writing that it has a no-training, zero-retention agreement covering our inputs and outputs, and that Section 13.3 does not permit using our identifiable content to build or improve AI features?

Addressed, but falls short of the standard

The Terms give a 90-day window to retrieve content after termination, but no document says when data is deleted after that, how backups are handled, or offers a certificate of destruction. The Terms reserve a right to archive data under unspecified 'standard policies.' The Privacy Policy (last updated January 2021) sets no retention periods.

• “Following any Cancellation or Termination of the Clearbrief Services for any reason, a User shall have ninety days to retrieve any and all Content.” Clearbrief Terms of Service, Section 7.4
• “except for Clearbrief's right to archive such User Provided Information in accordance with its standard policies” Clearbrief Terms of Service, Section 13.5

Ask the vendor: How many days after we cancel or request deletion will our documents and backups be permanently deleted, and will you confirm deletion in writing?

Clearbrief claims SOC 2 Type 2 on its homepage. Under the amended rule, that claim is accepted as current and covering the product, so the SOC 2 half now passes. The DPA half still fails. No Data Processing Agreement is published or referenced in the Terms or Privacy Policy, and there is no Trust Center where one could be requested (the /security page returns 404 and trust.clearbrief.com does not resolve). Clearbrief also markets medical timelines, but no BAA is mentioned anywhere. Rule update: a DPA is no longer required to pass. If a vendor has one, it should make it available on request, so a firm should ask for it before signing.

• “SOC 2, Type 2 Certified, with comprehensive data hygiene controls” Clearbrief homepage, Security section

Not found in public documents

Data location (United States) is published. No subprocessor list is published; the Privacy Policy refers only to unnamed 'agents and others' and a 'third-party cloud storage provider.' No commitment to give notice before adding subprocessors was found.

• “The Sites are hosted on servers located in the United States.” Clearbrief Privacy Policy
• “The Clearbrief Services are controlled and operated from our facilities in the United States.” Clearbrief Terms of Service, Section 19

Ask the vendor: Can you send your current list of subprocessors, including the AI provider and cloud host, and commit to notifying us before adding new ones?

Not found in public documents

Under the amended rule, the SOC 2 Type 2 claim would count as independent support. However, no official source states that one customer's data is isolated or logically separated from another's, so there is nothing for the audit to support. This remains a Fail.

Ask the vendor: How is our firm's data separated from other customers' data, including inside the AI features, and does your SOC 2 Type 2 audit test that separation?

Not found in public documents

Neither the Terms of Service nor the Privacy Policy contains any commitment to notify customers of a data breach, with or without a time frame. The only security contact published is hello@clearbrief.ai via the 'Talk to our security team' link.

Ask the vendor: Will you commit in writing to notify us of any security incident affecting our data within 72 hours, and to whom?

Addressed, but falls short of the standard

The Terms promise the ability to download previously input data on termination (format not specified), with a 90-day retrieval window. No document states that Clearbrief deletes or destroys the data afterward, and the Terms reserve a right to archive it.

• “Clearbrief will provide the ability for the User to download all data that the User previously input into their account” Clearbrief Terms of Service, Section 9

Ask the vendor: After we export our data, will you delete all copies, including backups, and provide written confirmation of destruction?

Nothing public describes limits on staff access, permission requirements, logging, or revocation. The Terms list broad grounds on which Clearbrief may access account contents, including to respond to customer service requests, and the Privacy Policy says Clearbrief may access data to improve services and test features.

• “Clearbrief may access, preserve and disclose your account information and related contents if required to do so by law” Clearbrief Terms of Service, Section 7.7

Ask the vendor: Which Clearbrief staff can see our documents, do they need our permission each time, and is that access logged and revocable?

No public source describes a log of AI prompts and responses or any customer-exportable audit log. Additional user guides exist but require signing in.

Ask the vendor: Can we export a log of every AI prompt and response, showing which user ran it?

Clearbrief works as a drafting and fact-checking add-in inside Microsoft Word, and nothing published describes it sending messages, filing, or acting on its own. The vendor says in general terms that customers control whether and how generative AI is used, but no document spells out approval steps or a stop control for specific features.

• “Clearbrief gives you control over if and how Generative AI is used” Clearbrief homepage, Security section

Ask the vendor: Can an administrator turn off generative AI features firm-wide, and does every AI output require a user to accept it before it goes into a document?

Clio's official subprocessor list names the AI providers used for Clio Grow: AWS Bedrock, Anthropic, OpenAI, Google Vertex AI, and Microsoft Azure Foundry. It does not say which model runs which Grow AI feature. The Terms of Service promise 30 days' notice before any change that materially reduces what the service does. That is a general change-notice clause, so it meets the bar, but it does not specifically cover switching AI models.

• “Anthropic, Inc: Clio Manage, Clio Grow. OpenAI, LLC: Clio Manage, Clio Grow.” Subprocessors (Exhibit C), last modified Aug 20, 2026
• “Themis shall provide Subscriber with 30-days notice of any modification that materially reduces the functionality of the Service.” North American Terms of Service, section 2.11

Clio Grow is sold through 'Book a demo' as an add-on or in the Elite plan, so a sales call comes first. Clio's help center shows Grow can be billed on a 30-day cycle, and the pricing page says monthly terms can be cancelled at the end of the term. Grow AI also has a free trial measured in hired matters rather than days. A firm can therefore get hands-on use on a month-to-month basis without an annual contract. Nothing published confirms a free trial of the base Clio Grow CRM itself.

• “If you have a 30-day billing cycle, the first invoice is paid in full.” Help: Account Management in Clio Grow
• “Cancellation takes effect at the end of your current term, whether monthly or annual.” Pricing page FAQ
• “New firms start with a free trial covering a number of hired matters” Help: Get Started with Grow AI

Both halves are covered. Clio's Security page says your data is never used to train AI models, and the Trust Centre says Clio has zero-data-retention agreements with its AI providers. The Terms of Service let Clio use only de-identified and aggregated AI inputs and outputs to improve its AI services, which is acceptable under the amended rule. Something to be aware of: the marketing says 'never,' while the contract allows that de-identified use. The subprocessor list also includes Braintrust, an AI evaluation provider used for Clio Grow.

• “Data is never used to train AI models or for any other external purposes.” Clio Security page
• “For AI subprocessors, we also implement Zero Data Retention (ZDR) agreements to further protect customer data.” Clio Trust Centre, Subprocessors item
• “Themis may de-identify and aggregate the Content submitted to, and Output received from, AI Services and use the same in order to improve” North American Terms of Service, section 14.6
• “does not grant Themis the right to use any of Subscriber's Confidential Information for the purpose of training generalized LLMs.” North American Terms of Service, section 14.6

Addressed, but falls short of the standard

The Terms say data is deleted after 'no less than' 90 days. That sets the earliest date, not a deadline. The firmer 90-day deletion promise is in the Data Protection Addendum, which applies only to EEA/UK data and has a legal-retention exception. Nothing public says when backups are purged. The Trust Centre lists a 'Certificates of Destruction' item, but its terms are not published. No public document names a specific, verifiable deletion mechanism for a US firm.

• “Following a period of no less than ninety (90) days from the cancellation or termination of a subscription, all Content associated with such subscription will be irrevocably deleted” North American Terms of Service, section 10
• “Themis and its Subprocessors may retain Subscriber Personal Data to the extent required by applicable laws” Themis Data Protection Addendum, section 7 (EEA/UK data only)
• “Policies, processes, and procedures for enabling data deletion are established and in place” Clio Trust Centre, Data Erasure item

Ask the vendor: After a US firm cancels Clio Grow, by what date will its data, including backups and copies held by AI subprocessors, be deleted, and will you provide a certificate of destruction on request?

Clio claims SOC 2 Type II on the Trust Centre page that covers Clio Grow. Under the amended rule, that audit is assumed current even though the last publicly posted period ended May 31, 2025. A Data Processing Agreement is listed in the Trust Centre's Legal section and can be requested before signing. The DPA published on the website covers only EEA/UK data. Clio Grow is an intake and CRM tool, not a medical records or PI file product. Clio offers a BAA with its HIPAA or Personal Injury add-on. Firms taking injury details from leads should confirm the BAA covers Grow.

• “our 2025 SOC 2 Type II report has been issued for the period of June 1st, 2024 to May 31st, 2025.” Clio Trust Centre, Compliance section and Trust Centre Updates
• “The Clio page covers Clio Manage and Clio Grow.” Help: Access Clio's Trust Center
• “Legal: Subprocessors, Data Processing Agreement, Data Subject Requests” Clio Trust Centre, Legal section (opened 2026-10-06)
• “you can purchase Clio's HIPAA Add-on or Clio's Personal Injury Add-On. Clio can then enter into a business associate agreement (BAA)” Help: Compliance

Addressed, but falls short of the standard

Clio publishes a detailed subprocessor list with the products each one serves and where data is stored (for example, Vapi for the Grow voice agent in the US, plus AI providers). The only written promise to notify customers before adding a subprocessor, with time to object, is in the DPA, and the DPA applies only to EEA/UK data. The US Terms of Service and Privacy Policy contain no advance-notice commitment. Two of the three parts are met. Notice before new subprocessors are added is missing for a US firm.

• “Vapi Inc | Clio Grow | AI enabled Voice Agent | United States” Subprocessors (Exhibit C)
• “Themis shall notify Subscriber from time to time of the identity of any Subprocessors engaged.” Themis Data Protection Addendum, section 3.2 (EEA/UK data only)

Ask the vendor: Will Clio commit in writing to notify our US firm before adding any new subprocessor for Clio Grow, and give us a chance to object?

Clio states clearly that it keeps each customer's data logically separate. Independent support comes from its claimed SOC 2 Type II audit, which is treated as current under the amended rule. Clio also says an outside cybersecurity firm tests its platform at least once a year, and the Trust Centre lists a Pentest Report Summary.

• “Clio logically separates each customer's data, ensuring only you and your team have access to your firm's information.” Clio Security page
• “At least annually, a leading cybersecurity firm conducts penetration tests on our platform” Clio Security page
• “our 2025 SOC 2 Type II report has been issued” Clio Trust Centre, Trust Centre Updates

The US Terms of Service commit Clio to report a security breach involving your content within 72 hours of learning of it. The Trust Centre also lists a Data Breach Notifications item. Public contact channels include privacy@clio.com for data requests and the Trust Centre support and vulnerability links.

• “Themis shall make such report within 72 hours after learning of the Security Breach.” North American Terms of Service, section 5.5

Clio Grow lets you export contacts and matters as CSV files. The Terms give you 90 days after cancellation to retrieve your content and say it will then be permanently deleted. Both parts are met. Note that 'no less than 90 days' sets no outer deadline, and proof of deletion is not described publicly (see Q4).

• “You can back up your Clio Grow data by exporting a CSV list of your contacts and matters.” Help: Filter and Export Matters in Clio Grow
• “Subscriber shall have ninety days to retrieve any and all Content.” North American Terms of Service, section 8.2
• “all Content associated with such subscription will be irrevocably deleted from the Service.” North American Terms of Service, section 10

Clio's Security page says support staff need your explicit permission in the app before accessing your data, and that the access is time-limited, logged, and revocable. All three elements for the top score are stated publicly.

• “If Clio support needs temporary data access, your explicit permission is required in the app. This access is time-limited, logged, and can be revoked” Clio Security page, Customer Support (CS) security

Clio says firms can see a log of every action taken by AI, and Grow AI has an Activity history of the individual actions it took. Grow AI replies are also filed against the related contact or matter. Nothing public confirms that every AI prompt and answer is logged in full or that the log can be exported.

• “You can also see a log of every action taken by AI within the firm.” Clio Security page
• “Activity history: A record of the individual actions Grow AI took.” Help: Manage Your Grow AI Subscription

Ask the vendor: Can we export a complete log of every Grow AI conversation, prompt, and response, tied to the user or lead, and how long is it kept?

Grow AI works on its own without approval of each message. It replies to prospective clients by email, chat, and phone, and books consultations for qualifying leads. Clio's help center documents what it does on its own and gives administrators ways to stop it: a firm-wide off switch and a Deactivate control that stops new actions across all channels. Uncertain emails are routed to staff for review.

• “Firm administrators can turn off Grow AI for the entire firm. Turning this off immediately removes Grow AI for every user” Help: Get Started with Grow AI
• “routes anything that is is unsure about to your lead inbox marked For review, allowing your team to decide how to handle it.” Help: Grow AI's Email Agent
• “Deactivating stops Grow AI from taking new actions across all channels.” Help: Manage Your Grow AI Subscription

Clio's published subprocessor list names the AI providers used for Clio Manage: AWS Bedrock, Anthropic, OpenAI, Google Vertex AI and Microsoft Azure Foundry. The list does not say which model handles which feature. The Terms of Service promise 30 days' notice before any change that materially reduces functionality, and advance notice of material changes to the contract.

• “AI Service Providers ... Anthropic, Inc ... OpenAI, LLC ... Google LLC (Google Vertex AI) ... Microsoft Corporation (Azure Foundry) ... Clio Manage” Subprocessors
• “Themis shall provide Subscriber with 30-days notice of any modification that materially reduces the functionality of the Service.” North American Terms of Service, s. 2.11
• “In the event of material changes to the Agreement, Themis will notify Subscribers, by email, or by other reasonable means” North American Terms of Service, s. 21.7

Clio offers a self-serve 7-day free trial of Clio Manage with no credit card, and plans can be billed monthly and cancelled at the end of the month. AI is included on the Core plan and above, so a firm can test the AI on a month-to-month Core plan without an annual contract. The pricing page does not say whether the free trial itself includes Manage AI.

• “Start your 7-day trial of Clio Manage. No credit card required.” Free trial signup page
• “Cancellation takes effect at the end of your current term, whether monthly or annual.” Pricing FAQ
• “Yes, on Core and above.” Pricing FAQ: Is AI included in my plan?

Clio's security page and pricing FAQ state, without qualification, that firm data is never used to train AI models, which covers both Clio's own models and the outside AI providers it uses (Anthropic, OpenAI, Google, Microsoft, AWS). The security page adds that its AI tools process data in real time without storing or reusing it. The Terms of Service (s. 14.6) bar Clio from using a firm's confidential information to train general AI models. The same section lets Clio use de-identified, aggregated AI inputs and outputs to improve its AI services. Under the amended rule that use is acceptable, but firms should know the contract allows it. No public contract clause names a no-training agreement with each AI provider, so this rests on Clio's official public statements.

• “Data is never used to train AI models or for any other external purposes.” Clio Security page
• “Your firm's data stays in Clio and is never used for AI training or any other external purpose.” Pricing FAQ: Is my firm's data used to train AI?
• “does not grant Themis the right to use any of Subscriber's Confidential Information for the purpose of training generalized LLMs.” North American Terms of Service, s. 14.6
• “Themis may de-identify and aggregate the Content submitted to, and Output received from, AI Services and use the same in order to improve” North American Terms of Service, s. 14.6

Addressed, but falls short of the standard

The contract sets a 90-day minimum before deletion after cancellation, but no deadline by which deletion must happen. No published source says when backups are purged or offers a certificate of destruction. The only firm deletion deadline (within 90 days of termination) is in the DPA, and the DPA applies only to EEA/UK customers. A US firm gets no specific deletion deadline or proof of deletion in public documents.

• “Following a period of no less than ninety (90) days from the cancellation or termination of a subscription, all Content ... will be irrevocably deleted” North American Terms of Service, s. 10.7
• “the Subscriber Personal Data relates to individuals located in the EEA; or (ii) Subscriber is established in the EEA or UK” Themis Data Protection Addendum (scope)

Ask the vendor: After we cancel, by what date will you delete our data, including all backups and AI provider copies, and will you give us a certificate of destruction on request?

Clio says it completes annual SOC 2 Type II audits, and its Trust Centre lists SOC 2 Type 2. Under the amended rule this is treated as current, even though the newest publicly announced audit period ended May 31, 2025. The published DPA on clio.com applies only to EEA/UK customers, but the Trust Centre lists a DPA document that can be requested before signing, which counts as available. For health information, Clio's help center says US firms that buy the HIPAA Add-on or Personal Injury Add-on can accept a BAA in Clio Manage. Clio will not accept edits to its BAA.

• “Clio successfully completes annual SOC 2 Type II and SOC 1 Type II examinations.” Clio Security page
• “Our SOC 1 and SOC 2 reports are available via trust.clio.com.” Clio Security page
• “you can purchase Clio's HIPAA Add-on or Clio's Personal Injury Add-On. Clio can then enter into a business associate agreement (BAA) with your firm” Compliance (help center)
• “Subscriber Personal Data relates to individuals located in the EEA; or (ii) Subscriber is established in the EEA or UK” Themis Data Protection Addendum

Addressed, but falls short of the standard

Clio publishes a detailed subprocessor list with data locations (US, EU, Canada, Australia), and the security page describes regional hosting. The only promise to notify customers about new subprocessors, with a right to object, is in the DPA, and the DPA applies only to EEA/UK customers. The US Terms of Service and the subprocessor page contain no such commitment. A US firm therefore has no published right to notice before a subprocessor is added.

• “Anthropic, Inc ... Clio Manage ... AI Processor ... United States” Subprocessors
• “Themis shall notify Subscriber from time to time of the identity of any Subprocessors engaged.” Themis Data Protection Addendum, s. 3.2

Ask the vendor: Will you commit in writing to notify us, and give us a chance to object, before adding any new subprocessor that will handle our data, including AI providers?

Clio states that each customer's data is logically separated. Its claimed annual SOC 2 Type II audit counts as current independent support under the amended rule. Clio also says an outside cybersecurity firm runs penetration tests at least once a year, though no summary of results is public.

• “Clio logically separates each customer's data, ensuring only you and your team have access to your firm's information.” Clio Security page
• “Clio successfully completes annual SOC 2 Type II and SOC 1 Type II examinations.” Clio Security page
• “At least annually, a leading cybersecurity firm conducts penetration tests on our platform” Clio Security page

The Terms of Service commit Clio to report a security breach affecting firm content within 72 hours of learning of it, with relevant details. No named breach contact is published. General contact is through support or the Trust Centre.

• “Themis shall make such report within 72 hours after learning of the Security Breach.” North American Terms of Service, s. 5.5

Clio says firms can export their data in open, nonproprietary formats while active or within 90 days of cancellation. The contract says content is irrevocably deleted after the retrieval period. Q4 notes that no deletion deadline or certificate is publicly committed.

• “retrieve your firm's data ... in open, nonproprietary formats at any time while your account is active, or within 90 days of cancellation.” Pricing FAQ: Can I export my data?
• “all Content associated with such subscription will be irrevocably deleted from the Service.” North American Terms of Service, s. 10.7

Clio's security page says support staff need the firm's explicit in-app permission to access data, and that access is time-limited, logged and revocable.

• “If Clio support needs temporary data access, your explicit permission is required in the app. This access is time-limited, logged, and can be revoked” Clio Security page

Clio Manage has an AI Event Log showing a timestamp, the user the AI acted for, and an event description, covering the last 90 days. Nothing public says it captures the full text of every prompt and answer, or that it can be exported.

• “The AI Event Log in your firm's Settings provides a record of your AI actions in Clio over the last 90 days” AI Features in Clio Manage
• “You can also see a log of every action taken by AI within the firm.” Clio Security page

Ask the vendor: Does the AI Event Log record the full text of every prompt and AI answer by user, can we export it, and can we keep it longer than 90 days?

Clio says users review and approve all AI output before it reaches a client or court. The help center describes AI-generated bills as ready for the firm's approval. Firm admins can switch off AI for all users at once. Some features, such as matter catch-up summaries, run in the background, but those stay inside the firm's own account.

• “you review and approve everything before it reaches a client or the court.” Pricing FAQ
• “Turning this off removes AI features for every user immediately” AI Features in Clio Manage

Vendor's own documents contradict each other

Darrow's terms name Anthropic (Claude) and OpenAI as model providers for Torch, and OpenAI for the Intelligence services. But the same terms say Darrow can switch AI providers at any time without prior notice. A general clause promises notice of material changes, so the documents point in two directions, and the specific language on switching models says no notice.

• “provided by the third-parties including Anthropic (Claude) and OpenAI (the "AI Service Providers")” Darrow Terms of Use, Part 2 Torch Terms
• “Company reserves the right to use different AI Service Providers at any time without providing prior notice.” Darrow Terms of Use, Parts 2 and 3
• “In case of a material change, Company shall notify the Customer by posting a notice in the Torch Service or on the Company's website” Darrow Terms of Use, Part 2 Torch Terms

Ask the vendor: Will you commit in writing to notify us before you change or add the AI model providers that process our firm's data?

Darrow publishes terms for a two-week, no-fee trial of its Drug Intelligence and Medical Device Intelligence services, which fit PI and mass tort work, with no payment obligation unless the parties later agree in writing. Torch trials are also mentioned, with terms set in an Order Form. The website itself only shows Book a Demo, so access appears to go through sales first, which is acceptable here.

• “made available to the Customer solely for evaluation purposes during a limited, no-fee trial period of two weeks” Darrow Terms of Use, Part 3 Intelligence Trial Services
• “The Company may offer, from time to time, some or all of the Torch Service on free trial versions” Darrow Terms of Use, Part 2 Torch Terms

Vendor's own documents contradict each other

The amended rule allows use of anonymized or aggregated data, but that is not what Darrow's license covers. Its terms grant Darrow a perpetual license to use the Customer Data itself, expressly including training foundation models, and that grant does not say the data is anonymized first. A separate clause on anonymous, aggregated information exists, but it does not limit the training license. Nothing published promises the AI providers will not train on your data either.

• “for the purpose of maintaining, developing including training foundation models and providing the Torch Service” Darrow Terms of Use, Parts 2 and 3
• “improve the functionality of the Service, and to train foundation models of the Service” Darrow Privacy Policy
• “Customer's use of AI Features is also governed by the AI Service Providers terms” Darrow Terms of Use, Part 2 Torch Terms

Ask the vendor: Will you remove the training license from your terms and sign a written commitment that our identifiable prompts, uploads, and outputs will never train your models or any AI provider's models?

Addressed, but falls short of the standard

Retention is described only as long as reasonably necessary. No deletion timeline, backup schedule, or certificate of destruction is published. The terms say Darrow may delete data after termination, not that it will.

• “for the time period reasonably necessary to achieve the purposes described in this Policy” Darrow Privacy Policy
• “Company may delete the Customer Data without retaining any copy thereof” Darrow Terms of Use, Part 2 Torch Terms

Ask the vendor: Within how many days of our request or contract end do you delete our data, including backups, and will you provide written confirmation?

Not found in public documents

The benefit of the doubt only applies when a vendor claims SOC 2, and Darrow makes no SOC 2 claim anywhere. There is no Trust Center or security page (darrow.ai/security returns not found), and the terms mention only reasonable industry-standard security measures. No DPA or BAA is published or listed as requestable. The Torch terms tell customers not to submit protected health information, which matters for PI firms handling medical records.

Ask the vendor: Do you have a SOC 2 Type II report covering this product, and can you provide it along with your DPA and a BAA before we sign?

Addressed, but falls short of the standard

Data locations are listed broadly (EEA, US, Israel, and others). Anthropic and OpenAI are named as AI providers, but no full subprocessor list is published and no notice before adding subprocessors is promised.

• “the Torch Service is hosted and made available by certain sub-processors of the Company” Darrow Terms of Use, Part 2 Torch Terms
• “including the European Economic Area, the United States, Israel” Darrow Privacy Policy

Ask the vendor: Can you send your full subprocessor list and commit to notifying us before adding a new one?

Not found in public documents

Darrow publishes no statement that one customer's data is kept separate from another's. It also makes no SOC 2 claim and publishes no penetration test or red-team exercise, so the amended rule has nothing to give credit for.

Ask the vendor: How do you keep our firm's data separated from other customers, and what independent audit or penetration test confirms it?

Not found in public documents

Neither the Terms nor the Privacy Policy includes a breach notification timeline. The only contact published is general (contact@darrow.ai).

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a breach involving our data, and who is the contact?

Addressed, but falls short of the standard

The terms put export on the customer but do not describe an export feature or format. Deletion after termination is optional ("may delete"), with no proof of destruction.

• “Customer is solely responsible to export all available Customer Data prior to such termination or expiration” Darrow Terms of Use, Part 2 Torch Terms

Ask the vendor: In what file format can we export all our data, and will you certify deletion afterward?

The Privacy Policy says employee access is limited to need-to-know. It does not say access requires your permission, is logged, or can be revoked.

• “We limit access of your Personal Information only to those employees, third party service providers or partners on a 'need to know' basis” Darrow Privacy Policy

Ask the vendor: Do your staff need our permission to access our data, and is that access logged and revocable?

Torch keeps chat history that users can view and delete, but nothing published describes an exportable log of every AI prompt and answer tied to a user.

Ask the vendor: Can we export a complete log of every AI prompt and answer, showing which user made it?

Darrow produces case leads and research for attorneys to review, and does not appear to act on its own. The FAQ says AI-generated insights are reviewed by legal professionals, and the terms make the customer responsible for checking output, but there is no specific approval gate or stop control described.

• “THE CUSTOMER IS RESPONSIBLE FOR EVALUATING THE RESULTS OF ITS USE OF THE AI FEATURES.” Darrow Terms of Use, Part 2 Torch Terms
• “AI-generated insights are reviewed by experienced legal professionals and domain specialists to ensure accuracy, legal relevance, and alignment with client needs.” Darrow FAQ

Ask the vendor: Does any Darrow feature act without a person approving it first, and how would we turn it off?

DISCO's own subprocessor list names OpenAI and Writer as its large language model providers. The DPA promises advance notice before a new subprocessor (such as a new AI provider) is added, with 14 days to object. The general terms say product changes on subscriptions take effect at the next renewal. The list does not say which model powers which Cecilia feature.

• “OpenAI, L.L.C. ... Large language model ... US, EU” DISCO Sub-Processor List
• “DISCO shall give Customer prior notice of the appointment of any proposed Sub-Processor by updating the Sub-Processor Site” Data Processing Addendum
• “DISCO may, from time to time, add, update, modify, rebrand or cease commercializing any Technology Offerings or features” Terms and Conditions

Not found in public documents

The pricing page sends buyers to a quote and demo request. It does not describe a free trial, sandbox, or self-use demo account. The Terms and Conditions mention a "Free Trial Service" and a transactional (non-annual) order type, but no official page says a prospective customer can get one.

• “Contact us for a quote and personalized demo” Ediscovery Pricing

Ask the vendor: Can our firm get hands-on access to DISCO and Cecilia on our own test data before signing an NDA or an annual subscription, and on what terms?

Addressed, but falls short of the standard

This still fails under the relaxed rule. DISCO's AI terms give it a license to use the customer content itself to improve its AI services. Nothing limits that to de-identified, anonymized, or aggregated data. We rechecked the AI terms on 2026-10-06 and found no wording about anonymization. No DISCO document says customer data stays out of training for DISCO's models or for its AI providers, OpenAI and Writer. Nothing describes a no-training or zero-retention agreement with those providers.

• “create derivative works of, and distribute Customer Content ... for purposes of providing and improving the AI Services” AI Supplementary Terms
• “the rights in Customer Data necessary for DISCO to provide and improve the DISCO Offerings” Terms and Conditions

Ask the vendor: Will you put in writing that our documents, prompts, and Cecilia answers, in identifiable form, are never used to train or improve DISCO's models or OpenAI's or Writer's models, and that those providers keep none of it?

Addressed, but falls short of the standard

DISCO commits to deleting data after the contract ends, but only according to its "standard practices." No number of days is given and nothing is said about backups. The Trust Center lists "Certificates of Destruction" as a control heading, but the details sit behind an access request, so no public document says a customer can get one.

• “all Customer Data in such Technology Offerings will be deleted from DISCO's systems in accordance with DISCO's standard practices” Terms and Conditions
• “except to the extent DISCO or its Affiliates are required to retain such Customer Personal Data under applicable laws” Data Processing Addendum

Ask the vendor: How many days after termination are our data and backups deleted, and will you give us a written certificate of destruction?

DISCO says it has an annual SOC 2 Type 2 audit. Under the amended rule we treat that audit as current and as covering the product. The report can be requested through the Trust Center. The DPA is published on DISCO's website and is also listed in the Trust Center. DISCO is a general ediscovery product, not a health-records product, so no BAA was required.

• “DISCO undergoes an annual SOC 2 Type 2 audit” Ediscovery Data Security features page
• “SOC 2 Report ... SOC 3 Report ... Data Processing Agreement” DISCO Trust Center
• “audit reports or certifications performed no more than twelve (12) months prior to Customer's audit request” Data Processing Addendum

DISCO publishes a list of its subprocessors with each one's purpose and location (including OpenAI, Writer, and Pinecone). The DPA promises advance notice of new ones through a Trust Center subscription, with 14 days to object. Locations are listed for each subprocessor.

• “Amazon Web Services ... Cloud service hosting provider ... US, Ireland, Canada” DISCO Sub-Processor List
• “prior notice of the appointment of any proposed Sub-Processor ... via subscription to DISCO's Trust Center” Data Processing Addendum

The security appendix of the DPA commits DISCO to keeping each customer's data logically separated. Two things back that up independently: DISCO's claimed SOC 2 Type 2 audit, which we treat as current, and its stated regular penetration testing. The isolation statement is short and does not mention Cecilia or the vector database used for AI search by name.

• “logical segregation of data, restricted (e.g. role-based) access and monitoring” Data Processing Addendum (Appendix 2 security measures)
• “DISCO performs regular vulnerability scans and penetration testing” Ediscovery Data Security features page

Addressed, but falls short of the standard

The DPA promises breach notice "without undue delay" but gives no number of hours. The Terms and Conditions do not set a deadline either. The Trust Center lists breach notification as a topic but the details are not public.

• “DISCO will notify Customer of such Personal Data Breach without undue delay unless prohibited by law” Data Processing Addendum

Ask the vendor: Will you commit in writing to notifying us of any breach involving our data within 72 hours, and who would contact us?

Addressed, but falls short of the standard

The DPA says DISCO will delete or return data, or give the customer a way to download it, at the end of the contract. It does not name an export format, and the deletion promise has an exception for data DISCO must retain. The general terms put the burden on the customer to retrieve data before termination. No published document promises both a usable export and confirmed destruction afterward.

• “DISCO will delete, return, or provide Customer with a mechanism to allow Customer to obtain a copy of or delete all Customer Personal Data” Data Processing Addendum

Ask the vendor: At the end of our contract, in what formats can we export all of our documents, coding, and Cecilia work product, and will you then certify in writing that everything was destroyed?

The DPA limits DISCO staff access to people who need it and requires periodic access reviews. The Trust Center lists access monitoring. Nothing says staff need the customer's permission before viewing its data, or that the customer can revoke that access.

• “granting access on a need-to-know basis, use of unique IDs and passwords for all users” Data Processing Addendum (Appendix 2 security measures)

Ask the vendor: Do DISCO employees need our explicit permission before they open our matter data, is each access logged, and can we see or revoke it?

The Trust Center lists "Audit Logging" as a general heading. We found no public statement that every Cecilia prompt and answer is logged by user or that customers can export that log.

Ask the vendor: Does DISCO keep a log of every Cecilia question and answer tied to the user who asked it, and can we export that log?

Cecilia answers questions with cited sources and builds summaries inside the platform. It does not send messages or file anything on its own. The AI terms make the customer responsible for reviewing outputs. No published description covers approval steps or a stop control for bulk features like automated document review.

• “It is Customer's responsibility to review and validate any Output.” AI Supplementary Terms

Ask the vendor: For Cecilia's automated review features, what can the AI tag or code without a reviewer approving it, and how do we pause or reverse a run?

Addressed, but falls short of the standard

The Trust Center mentions OpenAI and Anthropic only indirectly, in a sentence about not letting third-party AI providers train on your files. It does not say which model powers EvenUp or which features use which provider. The Subscription Terms let EvenUp add, change, or remove features with no notice requirement. The DPA promises notice before a new subprocessor is added, but that would not cover switching between providers already in use.

• “We do not allow third-party AI providers, including OpenAI and Anthropic, to use your case files to train their models.” EvenUp Trust Center FAQ
• “EvenUp reserves the right to incorporate new features in connection with the Service, and to modify or deprecate existing features” Subscription Terms and Conditions, Section 2.5
• “Prior to engaging any new Subprocessor that will process Customer Personal Data, EvenUp will provide Customer notice and an opportunity to object” EvenUp DPA For Customers, Section 4.1

Ask the vendor: Which AI models and providers power EvenUp's demand letters and other features, and will you notify us in writing before changing them?

Addressed, but falls short of the standard

The website offers only 'Schedule a Demo' and a sample demand letter. It has no free trial, pricing page, or self-serve signup. The Subscription Terms mention trial periods, but only inside a signed Order Form. That Order Form renews automatically at the full committed level unless you cancel in time.

• “an Order Form with a trial or ramp-up period will not renew with a trial or ramp-up period” Subscription Terms and Conditions, Section 4.1

Ask the vendor: Can our firm get hands-on use of EvenUp on a few real cases before signing an NDA or an Order Form that renews into a paid annual commitment?

Addressed, but falls short of the standard

The model-provider half is met: EvenUp says OpenAI, Anthropic, and other third-party AI providers may not train on your case files. EvenUp's own half is not. Under the amended rule, using de-identified or aggregated data would be fine, but EvenUp's terms do not limit its own use that way. Section 6.3 lets EvenUp use your inputs and outputs to improve its products, during and after the contract. The de-identification requirement in that section applies only when EvenUp shares data with others, not to its own use. The Trust Center also says customer cases help EvenUp's models recognize patterns. No published source promises that EvenUp's own models are not trained on your identifiable case content, and no opt-out is published. The Privacy Policy's AI training language does not decide this, because that policy says it does not cover data EvenUp processes for its business customers.

• “We do not allow third-party AI providers, including OpenAI and Anthropic, to use your case files to train their models.” EvenUp Trust Center FAQ
• “We leverage our experience handling thousands of cases to improve our product and help our models recognize patterns” EvenUp Trust Center FAQ
• “EvenUp has the right to collect, analyze, and aggregate Customer Input, Output” Subscription Terms and Conditions, Section 6.3
• “may during and after the term hereof (i) use such data to improve EvenUp's products and services” Subscription Terms and Conditions, Section 6.3

Ask the vendor: Will EvenUp agree in writing that its own models are trained or improved only on de-identified or aggregated data, and never on our firm's identifiable case files, inputs, or outputs, during or after our contract?

Addressed, but falls short of the standard

The Trust Center caps backup retention at up to one year. But neither the DPA nor the Trust Center gives a deadline for deleting live data after you ask. Both carve out retention allowed by 'EvenUp's policies' or for legal, compliance, security, audit, backup, or dispute reasons, and those exceptions are not defined. No certificate of destruction is offered.

• “Backup copies, where present, are retained in accordance with EvenUp's standard backup lifecycle and may be retained for up to one year before deletion.” EvenUp Trust Center FAQ
• “except where continued retention of Customer Personal Data is in accordance with applicable law or EvenUp's policies” EvenUp DPA For Customers, Section 3.3(g)

Ask the vendor: How many days after our request or contract end will EvenUp delete our case data from live systems, and will you provide written certification of deletion?

SOC 2 and the DPA now pass. EvenUp claims SOC 2 Type 2, and under the amended rule we assume that audit is current and covers the product. A DPA is published. The BAA is still the gap. EvenUp's core work involves medical records. The DPA excludes health information and says a separate BAA governs it 'where required or applicable', but no BAA is published. The Trust Center lists only a SOC 2 report, a HIPAA assessment, and CCPA documents as requestable, not a BAA. The Trust Center's BAA references describe agreements with EvenUp's own staff and service providers. Nothing public confirms a BAA is available to your firm before signing. Rule update: a BAA is required only when the product handles HIPAA-covered health information. Most plaintiff firms are not HIPAA covered entities, so this is not counted against the vendor. A firm that acts as a business associate of a healthcare provider should still ask for a BAA.

• “EvenUp's security and compliance program has been independently assessed through a SOC 2 Type 2 examination” EvenUp Trust Center
• “where required or applicable, will instead be governed by a separate Business Associate Agreement ("BAA") entered into by the Parties” EvenUp DPA For Customers, preamble
• “EvenUp personnel and service providers are subject to ... HIPAA-compliant business associate agreements.” EvenUp Trust Center FAQ, Who can access your case data?

Not found in public documents

Two of the three parts are published. Data location is U.S. data centers, and the DPA promises notice with a 10-business-day objection window before a new subprocessor is added. No subprocessor list is published. The DPA does not link to one, the Trust Center does not list one, and the Privacy Policy names only categories of service providers. OpenAI and Anthropic are mentioned only in passing as AI providers.

• “Case data is stored in U.S.-based data centers.” EvenUp Trust Center FAQ
• “Customer will have ten (10) business days from such notice to submit any objection in writing.” EvenUp DPA For Customers, Section 4.1

Ask the vendor: Can you send us your complete current subprocessor list, including every AI model provider and hosting provider, before we sign?

Both the Trust Center and the DPA say customer data is logically separated. EvenUp claims a SOC 2 Type 2 examination, which the amended rule treats as current independent support. The Trust Center also says a third party pen tests production at least once a year.

• “Our multi-tenant architecture uses logical separation, authenticated identities, role-based access, and scoped permissions” EvenUp Trust Center FAQ
• “Data security controls which include logical segregation of data, restricted (e.g., role-based) access” EvenUp DPA For Customers, Schedule 2
• “EvenUp's security and compliance program has been independently assessed through a SOC 2 Type 2 examination” EvenUp Trust Center

The DPA commits to notifying you of a security incident within 72 hours after EvenUp becomes aware of it. Notices to you go to the email on your Order Form, and you send legal notices to legal@evenuplaw.com.

• “without undue delay and in any event within seventy-two (72) hours after becoming aware” EvenUp DPA For Customers, Section 6.1

Not found in public documents

The DPA says EvenUp will delete or return personal data at termination, as you instruct, subject to the retention exceptions in Q4. No published source describes exporting your case data in a usable format such as PDF, CSV, or native files. The Subscription Terms say only that prepaid deliverables already requested will be delivered within 30 days of termination.

• “Upon termination of the Agreement, as instructed by Customer, delete or return Customer Personal Data” EvenUp DPA For Customers, Section 3.3(g)

Ask the vendor: At contract end, can we export all our case files, chronologies, and demand drafts in PDF or native formats, and will EvenUp then confirm in writing that all copies are destroyed?

EvenUp describes restricted, need-to-know staff access. Nothing published says staff need your permission to view your data, or that you can see or revoke staff access. EvenUp's model includes its own staff doing case work, so staff access is part of the service.

• “Access is limited to authorized EvenUp personnel and service providers on a need-to-know basis, with limited access permissions.” EvenUp Trust Center FAQ

Ask the vendor: Do EvenUp staff need our permission to open our case files, is every access logged, and can we see that log or revoke access?

EvenUp says case managers get summaries of what its communication agents found, which is a general form of AI activity oversight. Nothing published describes a log of every AI prompt and answer across the platform, tied to a user, and nothing says any log can be exported. EvenUp's agents guide lists transcripts and logging as things to look for in any agent, but does not say EvenUp's own agents provide them.

• “Case managers receive summaries showing which clients need personal follow-up” EvenUp Unveils Medical Management and AI Communication Agents

Ask the vendor: Can we export a complete log of every AI prompt, answer, and agent action in our account, showing which user or agent did it and when?

Demand letters and other deliverables are delivered as drafts for attorney review, which works as an approval gate. But EvenUp's communication agents contact clients on their own, and the Subscription Terms put the risk of these autonomous actions on the firm, reviewed or not. No published source says what the agents may do without approval or how a firm pauses or turns them off.

• “DELIVERABLES ARE PROVIDED IN DRAFT AND NOT FINAL FORM.” Subscription Terms and Conditions, Section 9.1
• “ACTIONS TAKEN AUTONOMOUSLY BY THE SERVICE ON CUSTOMER'S BEHALF ("AGENTIC ACTIONS"), WHETHER OR NOT INDIVIDUALLY REVIEWED BY CUSTOMER” Subscription Terms and Conditions, Section 9.3

Ask the vendor: Which agent actions (calls, texts, record requests) happen without our approval, and how do we pause or turn off an agent immediately?

Everlaw names OpenAI as its model provider, and OpenAI is on the published subprocessor list. The Terms promise email notice 30 days before material changes take effect, and the DPA promises 10 days notice before adding or replacing a subprocessor. Note that the Governance Framework says Everlaw keeps flexibility to shift between third-party providers, and AWS and Google are listed as 'Cloud and AI Service Provider', so OpenAI may not be the only model in use.

• “Everlaw uses OpenAI as our LLM vendor.” Everlaw's Commitment to Accuracy and Confidentiality for EverlawAI Assistant
• “Everlaw will notify Customer of material changes by sending an email to Customer's address and such changes will be effective 30 calendar days following notification” Everlaw Customer Terms of Service, Section 13.3
• “Everlaw will notify Customer if it adds or replaces any Sub-processors at least 10 days in advance of any such changes.” Everlaw Customer Data Processing Addendum, Section 11.1

Not found in public documents

The Terms contain a generic free trial clause, but no Everlaw page actually offers a trial, sandbox, or month-to-month plan. The pricing page only offers 'Request a Demo'. Third-party sites mention a free trial and a monthly starting price, but Everlaw's own pages do not confirm either.

• “FREE TRIAL SERVICES ARE PROVIDED 'AS IS' WITHOUT ANY EXPRESS OR IMPLIED WARRANTIES OR REPRESENTATIONS OF ANY KIND.” Everlaw Customer Terms of Service, Section 13.2

Ask the vendor: Can our firm get hands-on access to Everlaw, including the AI Assistant, through a free trial or month-to-month plan before signing an NDA or annual contract, and for how long?

Both halves are covered on official pages. Everlaw says it does not train or fine-tune any AI model, and says OpenAI and any other model provider it uses are barred from training on or keeping customer data (zero data retention). Something to be aware of: the contract wording limits the promise to models 'for general use', which is narrower than the website statements. Nothing published permits training on identifiable customer data.

• “Everlaw does not develop generative AI systems or LLMs independently. We also do not train or fine-tune LLMs developed by others.” EverlawAI Governance Framework
• “OpenAI is expressly prohibited from (1) training its models with this data or (2) retaining this data for any period after processing” Everlaw's Commitment to Accuracy and Confidentiality for EverlawAI Assistant
• “Everlaw only works with LLM providers that commit to zero data retention, and do not use Everlaw customer data to train their LLMs” EverlawAI Governance Framework
• “Everlaw will not use Input or Output to train or finetune a generative AI model for general use through the Service.” Everlaw Customer Terms of Service, Section 8.5

Addressed, but falls short of the standard

The DPA says ending the subscription is an instruction to delete all customer data including copies, with a 30-day window to export first, but deletion happens within 'a commercially reasonable timeframe' with an exception for legally required retention. No fixed number of days, backup deletion schedule, or certificate of destruction is published. The Trust Center lists 'Customer data deleted upon leaving' as a control with no timeframe.

• “within a commercially reasonable timeframe” Everlaw Customer Data Processing Addendum, Section 6

Ask the vendor: Within how many days after termination or a deletion request will Everlaw delete our data, including backups and AI-processing copies, and will you provide written certification of deletion?

Everlaw claims SOC 2 Type 2 (period March 1, 2025 to February 28, 2026, with a 2026 bridge letter), and the full report can be requested through the Trust Center. The DPA is published on the website. Because ediscovery files often include medical records, note that Everlaw's legal page links a public BAA request form and the Trust Center lists a HIPAA-HITECH assessment. Everlaw also holds ISO 27001, 27017, 27018, and FedRAMP Moderate.

• “following the SOC 2 Type 2 examination period ended February 28, 2026” Everlaw Trust Center, 2026 SOC 2 Bridge Letter
• “covering the period of March 1, 2025 to February 28, 2026” Everlaw Trust Center, Everlaw SOC 3
• “Customer may choose within which of Everlaw's available AWS instances to have its Customer Data stored and processed.” Everlaw Customer Data Processing Addendum
• “Request for Business Associate Agreement” Everlaw Legal index

A subprocessor list with purposes and locations is published (AWS, Google, OpenAI, Zendesk, Pinecone, Datadog, Everlaw affiliates), the DPA promises 10 days notice before adding or replacing one, and customers choose a hosting region (US, UK, Germany, Canada, Australia). One gap: endpoint security tool subprocessors are not listed online and are available only by emailing privacy@everlaw.com.

• “Everlaw may provide limited access to Customer Data to the following Sub-processors.” Everlaw List of Sub-processors
• “Everlaw will notify Customer if it adds or replaces any Sub-processors at least 10 days in advance of any such changes.” Everlaw Customer Data Processing Addendum, Section 11.1
• “Customer may choose within which of Everlaw's available AWS instances to have its Customer Data stored and processed” Everlaw Customer Data Processing Addendum, Section 10.1

The DPA states customer data is logically segregated in Everlaw's multi-tenant environment, and the AI commitment page says data is never shared between customers. This is backed by Everlaw's SOC 2 Type 2 report and an NCC Group penetration test whose executive summary is listed in the Trust Center.

• “Customer Data within Everlaw's multi-tenant environments is logically segregated and attempts to access Customer Data outside allowed domain boundaries are prevented and logged.” Everlaw Customer Data Processing Addendum, Schedule 3
• “Our systems are designed so that data is never shared between Everlaw customers.” Everlaw's Commitment to Accuracy and Confidentiality for EverlawAI Assistant
• “following the SOC 2 Type 2 examination period ended February 28, 2026” Everlaw Trust Center, 2026 SOC 2 Bridge Letter

Addressed, but falls short of the standard

The DPA promises notice 'without undue delay' after a verified security incident, with no number of hours. The Terms give security@everlaw.com as the channel for customers to report incidents but set no timeline for Everlaw's own notice.

• “notify Customer without undue delay” Everlaw Customer Data Processing Addendum, Section 7.2

Ask the vendor: Will Everlaw commit in writing to notify us within 72 hours of discovering a security incident affecting our data, and who will contact us?

Customers can export whole projects with native, image, and text files, review work, load files, and CSVs, and the DPA says Everlaw deletes all customer data, including copies, after termination. A certificate of destruction is not offered publicly, and the deletion timeline is undefined (see Q4).

• “All the documents in the project (native, image, and text)” Everlaw Help Center: Export a Project
• “serves as Customer's instruction to Everlaw to delete all Customer Data, including copies, still in its possession or control” Everlaw Customer Data Processing Addendum, Section 6.2

Staff access is role-based, temporary, tied to a verified customer support request, and logged. The published materials do not clearly say staff need the customer's explicit permission in every case, or that the customer can revoke access.

• “This procedure entails verification of the customer request and evaluation of the necessity for access” Everlaw Security page
• “Everlaw maintains audit logs of requests and approvals per our internal procedures.” Everlaw Security page

Ask the vendor: Do Everlaw staff need our explicit written permission before accessing our data, can we see the access logs, and can we revoke access at any time?

Everlaw says it logs every generative AI request with the input, output, prompt, user, time, and documents affected, and shows this in a user-facing dashboard. The page does not say customers can export this log; the 'fetched by developers' wording refers to Everlaw's own servers for auditing, not a customer export.

• “Whenever an Everlaw user requests a generative AI task, Everlaw logs the request, as well as the input, output, and prompt where applicable.” EverlawAI Governance Framework

Ask the vendor: Can our firm export a complete log of every AI prompt and answer, tied to each user, and in what file format?

EverlawAI tasks run when a user asks for them, results stay inside the review platform with source citations, and AI features can be turned on or off. Nothing published describes the AI sending messages or filing on its own. However, human review is described as encouraged, not as a required approval step before AI output (such as coding suggestions) is applied to the record.

• “Users can opt in and opt out of using Everlaw AI.” EverlawAI Governance Framework
• “we encourage users to perform human verification of all generative AI outputs” EverlawAI Governance Framework
• “Customers are in full control over if and when these features are used.” Everlaw's Commitment to Accuracy and Confidentiality for EverlawAI Assistant

Ask the vendor: Must a person approve AI-generated coding, descriptions, or drafts before they are applied to our documents or work product, and can an administrator turn off AI features for the whole firm?

Filevine's official subprocessor list names OpenAI, Anthropic, Google, and X.AI as AI processing providers. It does not say which model powers which feature (LOIS, AIFields, DemandsAI). The Subscription Agreement promises 30 days' notice of material updates, though updates for new functionality, non-material changes, and legally required changes are excluded. The DPA offers a subscription for notices of new subprocessors.

• “OpenAI, L.L.C. ... Anthropic PBC ... Artificial intelligence processing” Filevine Subprocessors
• “Filevine will provide at least thirty (30) days' notice before an update takes effect” Filevine Subscription Agreement
• “whereupon Filevine will provide notification of any new Sub-processors” Filevine Data Protection Agreement, Section 9.2

Filevine offers a free LOIS evaluation (90 days by default) under a click-through evaluation agreement, with no NDA or paid contract required. The trial is limited to active legal professionals. The evaluation covers LOIS Console only. Filevine case management and the full AI feature set (which would include AIFields and DemandsAI) need a paid Sales Order. The evaluation terms also bar uploading protected health information, so you cannot test medical-records features with real files.

• “The Evaluation Services are provided to Evaluator at no cost during the Evaluation Period.” Filevine LOIS Evaluation Agreement
• “if no Order Confirmation specifies a period, ninety (90) days from first access” Filevine LOIS Evaluation Agreement, Section 5.1
• “Try LOIS today for free. No commitment required.” LOIS free trial landing page

Both halves are covered. Section 4.3 of the Subscription Agreement bars Filevine's outside AI providers from training on your data. Filevine's own LOIS page says its AI is never trained on your data. The contract gives Filevine no right to use identifiable customer content to train AI. It only lets Filevine use de-identified, aggregated data to improve the service, which this scorecard accepts. Be aware the no-training promise for Filevine itself appears on a marketing page rather than in the contract.

• “under written agreements prohibiting those providers from using Subscriber's Data to train or improve their models” Filevine Subscription Agreement, Section 4.3
• “AI Never Trained On Your Data” LOIS demo landing page
• “Filevine may use Subscriber's Data in de-identified, aggregated form ("Anonymized Data") to improve and enhance the Services” Filevine Subscription Agreement, Section 4.4

The published documents name a specific check. Filevine will give a written certificate of destruction, and the DPA puts that certificate within 30 days of destruction. After termination you have 20 days to choose destruction or return, and Filevine does not have to keep data more than 30 days. No document gives a timeline for deleting backups.

• “Filevine will certify in writing that it has destroyed Personal Information within thirty (30) days after it completes the destruction.” Filevine Data Protection Agreement, Section 13.3
• “Filevine shall not be obligated to store any Data for more than thirty (30) days following the expiration or termination of this Agreement.” Filevine Security Addendum, Section 13

SOC 2 and the DPA now pass. Filevine claims SOC 2 Type II, and under the amended rules we assume the audit is current and covers LOIS. The DPA is published. This still fails on the BAA. Filevine sells AI for medical records and personal injury demand letters, which handle health information. No published BAA was found, and no official page says one is available before signing. The Medical Records AI page and the Security page mention HIPAA but not a BAA. Rule update: a BAA is required only when the product handles HIPAA-covered health information. Most plaintiff firms are not HIPAA covered entities, so this is not counted against the vendor. A firm that acts as a business associate of a healthcare provider should still ask for a BAA.

• “assessed by independent third-party auditors on at least an annual basis through a SOC 2 Type II audit” Filevine Security Addendum, Section 1.1
• “LOIS is SOC 2 Type II certified and HIPAA-aligned” LOIS free trial landing page
• “Uses industry-leading security measures, HIPAA compliance, and a 0-day data retention policy.” Medical Records AI product page

All three are published: a subprocessor list with addresses and purposes, notice of new subprocessors (you must subscribe by email) with a 10-business-day objection window, and US data location. A few listed subprocessors are outside the US (for example, Humanloop in the UK).

• “Anthropic PBC ... Artificial intelligence processing” Filevine Subprocessors
• “Filevine shall also provide a mechanism to subscribe to notifications of new Sub-processors” Filevine Data Protection Agreement, Section 9.2
• “Subscriber's Data is stored at rest in data centers located in the United States, unless otherwise specified in the applicable Sales Order.” Filevine Security Addendum, Section 2.1

Not found in public documents

Filevine's SOC 2 Type II claim now counts as current independent support. But none of the official documents reviewed says one customer's data is kept separate from another customer's. That includes the Security page, which was checked again. The only 'logical separation' language is about encryption key systems, not customers. With no isolation statement, the audit has nothing to back up.

• “logically separates encryption key management systems from Subscriber's Data” Filevine Security Addendum, Section 3.2

Ask the vendor: Can you show us in writing how our firm's data, prompts, and AI outputs are kept separate from other Filevine customers, and which audit or penetration test covers that control?

Addressed, but falls short of the standard

Both the DPA and the Security Addendum say 'without undue delay' but give no number of hours. Notice goes to the contact named in your Sales Order. The Security Addendum also starts the clock only when Filevine 'confirms' a breach.

• “Filevine will notify Subscriber without undue delay, at the contact identified in the applicable Sales Order.” Filevine Security Addendum, Section 8.1
• “Filevine will promptly notify Subscriber without undue delay after becoming aware of a Security Breach.” Filevine Data Protection Agreement, Section 7.1

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a security breach that affects our data?

Addressed, but falls short of the standard

The destruction half is covered, with a written certificate available. The export half falls short: the documents say data will be returned in a 'mutually agreed-upon format' and name no usable format such as CSV, PDF, or native files. No official page reviewed describes a self-service export of AI outputs. You also must choose return within 20 days of termination.

• “returned to Subscriber in a mutually agreed-upon format” Filevine Security Addendum, Section 13
• “Filevine will destroy or, if directed in writing by Subscriber, return all or any Personal Information” Filevine Data Protection Agreement, Section 13.1

Ask the vendor: In what file formats will you return all our case data, documents, and AI outputs at the end of the contract, and can we export them ourselves before termination?

Staff access is limited to support and legal needs, requires unique credentials and multi-factor login, and activity is logged. No document says Filevine needs your permission before staff view your data, or that you can revoke staff access.

• “Filevine personnel will not access Subscriber's Data except (a) to provide or support the Services, or (b) to comply with applicable law” Filevine Security Addendum, Section 4.2
• “with actions of individual users traceable to those users” Filevine Security Addendum, Section 9.1

Ask the vendor: Do Filevine staff need our explicit permission before viewing our data, and can we see and revoke that access?

Filevine keeps internal security audit logs for at least one year. No official source reviewed describes a log of AI prompts and answers that a customer can view or export.

• “Filevine maintains audit logs reasonably sufficient to monitor, investigate, and report unauthorized or inappropriate system activity” Filevine Security Addendum, Section 9.1

Ask the vendor: Can our firm export a complete log of every LOIS, AIFields, and DemandsAI prompt and answer, tied to the user who ran it?

Filevine's documents call for human review of AI output in general terms and put that duty on the customer. No official source lists what the AI can do without approval (for example, sending messages or populating fields automatically). None describes an admin control to turn AI features off or stop them.

• “Subscriber is solely responsible for reviewing, verifying, and determining whether and how to use any Output” Filevine Subscription Agreement
• “must be independently reviewed by a qualified professional before use” Filevine LOIS Evaluation Agreement, Section 1.4

Ask the vendor: Which Filevine AI features can act or write to case records without a person approving first, and how can an admin turn those features off?

Not found in public documents

No official Gideon source could be reached. The vendor's domain now shows an unrelated casino site, so nothing names the AI model or commits to change notice.

Ask the vendor: Which AI model or provider powers Gideon, and where is your written commitment to notify customers before material changes?

Not found in public documents

No official source shows a trial, pilot, or month-to-month option. The only listing found (Clio's app directory) points back to the defunct domain.

Ask the vendor: Can our firm test Gideon hands-on before signing an NDA or an annual contract?

Not found in public documents

No published statement could be found on whether inputs or outputs train Gideon's models or the underlying provider's models.

Ask the vendor: Do you put in writing that our inputs and outputs are never used to train your models or your AI provider's models?

Not found in public documents

No retention or deletion terms could be found. The Privacy Policy and Terms URLs now redirect to an unrelated casino site.

Ask the vendor: How many days after a request do you delete our data, including backups, and how do you confirm it?

Not found in public documents

No SOC 2 Type II report, Trust Center, or DPA could be found. Intake data can include sensitive personal and injury details, so a BAA may also matter.

Ask the vendor: Can you share a current SOC 2 Type II report covering Gideon and a DPA (and a BAA if needed) before we sign?

Not found in public documents

No subprocessor list, notice commitment, or data location statement could be found.

Ask the vendor: Where is your subprocessor list, how much notice do you give before adding one, and where is our data stored?

Not found in public documents

No statement on separation of customer data and no independent audit could be found.

Ask the vendor: How is our firm's data kept separate from other customers, and what independent audit confirms it?

Not found in public documents

No breach notification timeline could be found in any reachable document.

Ask the vendor: Will you commit in writing to notify us of a breach within 72 hours, and through what contact?

Not found in public documents

No published export or post-termination destruction terms could be found.

Ask the vendor: In what format can we export all our intake data, and will you confirm deletion in writing afterward?

Nothing public on vendor staff access to customer data.

Ask the vendor: Do your staff need our permission to access our data, and is that access logged and revocable?

Nothing official on logs of AI conversations. Clio's listing (third party) says chat conversations are saved as documents in Clio Manage, which could not be confirmed on a vendor source.

Ask the vendor: Can we export a complete log of every chatbot conversation, tied to the user and date?

Clio's listing (third party) describes Gideon chatbots qualifying and routing leads without human involvement. No official source on approval steps or a way to stop the AI could be found.

Ask the vendor: What can the chatbot say or do with prospective clients without our approval, and how do we switch it off?

Harvey's own subprocessor list names its AI model providers: Microsoft (Azure), OpenAI, Anthropic, Google, AWS, Mistral and others. Harvey uses several models, not one. The Platform Agreement promises at least 15 days' notice before updates to the terms, except updates Harvey judges non-material, new features, or legally required changes. The DPA promises 30 days' notice before a new subprocessor is added, but only if you subscribe to email alerts. No document promises notice before Harvey switches which existing model handles your work.

• “OpenAI, LLC: AI provider ... Anthropic, PBC: AI provider ... Google Cloud Platform: AI provider” Subprocessor List
• “Harvey will provide notice of any update at least 15 days in advance of the update being posted.” Platform Agreement, Section 1.3.1
• “Harvey will notify You if Harvey intends to add one or more Subprocessors to the Subprocessor List at least 30 days before the change takes effect.” Data Processing Addendum

Harvey publishes separate Evaluation Terms for an authorized free trial. You accept them by using the service, with no NDA or annual contract required. The trial must be approved by Harvey through its sales team. No trial length is published. The website shows no pricing page or self-serve signup, only a demo request. Some independent reviews say Harvey does not usually offer trials to small firms, so a small firm should confirm it qualifies.

• “You are utilizing the Service via an authorized free trial” Evaluation Terms of Service
• “Use of the Services for unpaid or evaluation purposes.” Support and Service Level Terms

The Platform Agreement says in writing that neither Harvey nor its subprocessors (which include the AI model providers) will train any AI model on your content or data. Both halves of the test are met, and the record shows no use of identifiable customer content for training. Something to be aware of: optional 'Covered AI Services', which an administrator must turn on, let model providers keep data for safety reviews. The no-training promise still applies to those features.

• “Harvey will not train any AI models using Your Content or Customer Data. Subprocessors will not train any AI models using Your Content or Customer Data.” Platform Agreement
• “requires its model providers to commit to the same “no-training on customer data” and zero-retention posture” Trust Center: AI Training Data and Bias
• “Subprocessors may retain Customer Data and Content to the extent necessary to provide their services, perform safety reviews, or comply with applicable law” Service Terms (Covered AI Services)

Harvey gives specific timeframes. Data is deleted within 30 days of termination or a reasonable request, and backups expire after about 30 days. Customers can also set their own retention periods. Two limits apply. The Service Terms say the retention setting does not apply to 'certain features' listed only in documentation that requires a login. Only the EU transfer terms mention certifying deletion.

• “Within 30 days of either (i) termination of the Service; or (ii) upon Your reasonable request; Harvey shall ... return to You or delete” Data Processing Addendum
• “Backups are taken at least daily and are generally retained for up to 30 days, after which they are automatically expired or securely deleted” Trust Center: Data Backups
• “Within 30 days of termination, Harvey will securely delete any remaining Customer Data or Content unless otherwise instructed by You.” Platform Agreement

Harvey's Trust Center announces a completed 2026 SOC 2 Type II audit, which is treated as current and covering the product. The DPA is posted publicly, and a BAA is offered before signing for customers handling health information. Harvey also holds ISO 27001, 27701 and 42001 certifications.

• “Announcing Successful Completion of 2026 SOC 2 Type II Audit and ISO 27001 Certification” Trust Center update
• “Harvey AI provides a standard online Data Processing Addendum” Trust Center: Data Processing Addendum (DPA)
• “Harvey AI offers a Business Associate Addendum (BAA) for customers subject to HIPAA who intend to use the Service to process Protected Health Information” Trust Center: Business Associate Addendum

All three are published. Harvey lists each subprocessor with its purpose and location, and gives 30 days' notice before adding one (you must subscribe to the email alerts). Data is stored in the region on your order form (US, EU, Switzerland or Australia). Some features, such as web browsing, Ask LexisNexis and Anthropic models, may process data outside your chosen region.

• “Microsoft Corp: AI provider and cloud services provider: US, EU, Switzerland, Australia” Subprocessor List
• “If You subscribe to receive email notifications at the Subprocessor List, then Harvey will notify You ... at least 30 days before the change takes effect.” Data Processing Addendum
• “stores and processes Customer Data and Content in the geographic region specified on a customer’s order form” Trust Center: Geographic Location of Data

Harvey states that each customer's data is logically separated. This is supported by its claimed SOC 2 Type II audit and by annual independent penetration tests. Pen test summaries are available on request.

• “Customer data for each customer is logically separated to prevent any commingling of data between customers” Harvey Security page
• “logically isolated customer environments. We undergo regular independent audits such as SOC 2 Type II and ISO/IEC 27001” Trust Center: Data Security
• “Harvey engages an independent third party to conduct penetration tests of the Service at least annually.” Security Addendum

The Security Addendum promises breach notice within 48 hours, well inside the 72-hour benchmark. The two documents start the clock differently. The contract says 48 hours after Harvey becomes aware, while the Trust Center says 48 hours from confirmation. The contract wording is the stronger of the two. Notice goes to the security email on your order form. Published contacts are trust@harvey.ai and privacy@harvey.ai.

• “Harvey will notify You without undue delay, and in any case, within 48 hours after becoming aware.” Security Addendum, Section 8.1
• “we will notify affected customers without undue delay and in any case within 48 hours of confirmation per our Security Addendum” Trust Center: Incident Response

Addressed, but falls short of the standard

Deletion is clearly promised: within 30 days of termination or request. The export side is not. No public document describes how a customer gets all its data back (documents, Vault contents, work product) in a usable format such as native files, PDF or CSV. The DPA says Harvey will 'return or delete', which leaves the choice open, and the Platform Agreement covers deletion but not retrieval. Admins can export query history. Help center articles on file export require a login.

• “Harvey shall, and shall direct each Subprocessor to, return to You or delete the Customer Personal Data” Data Processing Addendum
• “authorized admins can see user email, timestamp, query type, query, and response, export this data” Trust Center: Audit Logging

Ask the vendor: Before we terminate, can we export all of our Vault documents, work product, and history in native, PDF, or CSV format, and will you confirm deletion in writing afterward?

Staff access is restricted, time-limited, logged and reviewed at least quarterly. Harvey does not say staff need your permission before accessing your data, or that you can revoke their access.

• “Harvey AI personnel have no standing access to customer environments and must request time-bound, audited elevation for specific tasks” Trust Center: Data Access
• “Harvey personnel will not access Customer Data or Content except (i) to provide or support the Service or (ii) to comply with the law” Security Addendum

Ask the vendor: Will Harvey staff ever access our data without our explicit permission, and can we see and revoke that access?

Firm administrators can see and export a workspace history of queries and responses, tied to the user's email and a timestamp. Separate activity logs, which record actions but not content, are available through an API.

• “workspace query history (including content) in the admin dashboard, where authorized admins can see user email, timestamp, query type, query, and response, export this data” Trust Center: Audit Logging

For agents, Harvey describes approval steps: you approve the plan before work starts and review the work 'before it goes out.' Harvey also offers scheduled agents that run in the background. No public document lists what agents can do without approval or explains how to pause or stop a scheduled agent.

• “Preview the plan, adjust the scope, and approve work before Harvey begins.” Harvey Agents product page
• “automatically running reports, updates, and reviews on the schedule you choose” Harvey Agents product page

Ask the vendor: What can Harvey agents do without a person approving it, and how do we pause or stop a scheduled agent?

InQuery's own subprocessor list names OpenAI and Anthropic as its AI model providers. The DPA commits to 14 days' email notice before adding a new subprocessor, which would cover adding a new model provider. Product pages only say the product is AI-enabled and do not name the models, and no notice is promised for switching between the two listed providers.

• “OpenAI | All Services | United States | AI model inference and processing” Sub-processor List
• “Anthropic | All Services | United States | AI model inference and processing” Sub-processor List
• “InQuery will notify Customer of its intent to update the Subprocessor List at least 14 days prior to engaging a new Subprocessor” Data Processing Addendum, Section 6.2

Addressed, but falls short of the standard

After a demo, InQuery offers to process one real case for free so you can see the output. InQuery runs the product for you in that case, so your firm does not get hands-on use. Nothing published describes a self-use trial, sandbox, or month-to-month plan, and no pricing or contract terms are public.

• “Send us a real case file and we'll process it end-to-end for free” Get Started

Ask the vendor: Can our firm use InQuery hands-on ourselves on a real case before signing an NDA or an annual contract, and what are the contract length options?

Not found in public documents

InQuery itself promises not to train on customer data, so the vendor half now passes. Its DPA also lets it create de-identified data to improve its products, which the amended rule accepts. Be aware of that clause, but it no longer causes a fail. The other half is still missing. InQuery sends data to OpenAI and Anthropic, and no published document (rechecked 2026-10-06) says those providers are barred from training on it or that a zero-data-retention agreement is in place.

• “Customer data is not used for AI model training, sold, licensed, or processed for any secondary commercial purpose.” Security Policy
• “OpenAI | All Services | United States | AI model inference and processing” Sub-processor List

Ask the vendor: Do your agreements with OpenAI and Anthropic bar them from training on our data or keeping it (zero data retention), and will you put that commitment in our contract?

Both the security page and the DPA commit to deleting customer data and backups within 28 days of case completion, with an automatic deletion certificate within 48 hours. The DPA defines case completion as formal closure with no pending actions, appeals, or holds, and carves out data that law requires InQuery to keep.

• “Customer data is deleted within 28 days of case completion, with backup copies purged within the same window” Security Policy
• “a deletion certificate will be automatically generated and provided to Customer within 48 hours” Data Processing Addendum

Addressed, but falls short of the standard

The benefit of the doubt does not help here, because InQuery's own security page claims only SOC 2 Type I and says Type II is still in progress. The amended rule still fails Type I alone. A DPA is published. The product handles medical records, but InQuery mentions BAAs only with its own subcontractors and confirms no BAA with customers before signing.

• “InQuery maintains SOC 2 Type I attestation from a qualified third-party auditor and is actively progressing toward SOC 2 Type II.” Security Policy
• “Business associate agreements ensure subcontractors comply with HIPAA privacy and security requirements.” Security Policy

Ask the vendor: When will your SOC 2 Type II audit be complete, and will you sign a BAA with our firm before we upload any client medical records?

A subprocessor list is published (AWS, Stripe, OpenAI, Anthropic), each with a United States processing location. The DPA gives 14 days' email notice before a new subprocessor and lets the customer object, with cancellation and a refund of prepaid fees as the remedy. The list has no last-updated date.

• “AWS | All Services | United States | Cloud infrastructure and storage” Sub-processor List
• “InQuery will notify Customer of its intent to update the Subprocessor List at least 14 days prior to engaging a new Subprocessor” Data Processing Addendum, Section 6.2

Addressed, but falls short of the standard

InQuery says customer data is isolated between customers. Its only SOC 2 claim is Type I, which the amended rule does not accept as independent support. It says it does annual penetration testing, but it publishes no report or summary, so nothing independent backs up the isolation claim.

• “Environment segmentation isolates customer data and prevents unauthorized cross-tenant access.” Security Policy
• “Annual penetration testing identifies vulnerabilities with high-risk findings tracked and remediated per policy.” Security Policy

Ask the vendor: Can you share a penetration test summary or other independent test showing that one customer's data is kept separate from another's?

Vendor's own documents contradict each other

The security page promises notice within 4 hours for incidents involving health data and within 72 hours for other confirmed incidents. The DPA, which is the contract document, repeats the 4-hour health data rule but says only 'without undue delay' for other incidents and gives no number. Because the two documents differ on non-health incidents, the 72-hour promise is not confirmed in the contract. Subprocessor objections go to privacy@inquery.com, but no breach contact is named.

• “within 72 hours for all other confirmed security incidents” Security Policy
• “InQuery will notify Customer without undue delay after becoming aware of a Security Incident.” Data Processing Addendum, Section 4.1

Ask the vendor: Will you put the 72-hour notice for all security incidents into the DPA itself, and who is our contact for a breach?

Addressed, but falls short of the standard

Deletion with a certificate is well documented. Export is described only in general terms ('exports', 'data portability'). No published source names the file formats you would get back, and the DPA says nothing about returning data, so the usable-format half is not confirmed.

• “Deliverables arrive as organized files and exports that drop into whatever your firm already uses” Plaintiff Firms page
• “Data portability requests are verified, logged, and fulfilled within one month using secure transfer mechanisms.” Security Policy

Ask the vendor: In what file formats can we export our chronologies, demands, and uploaded records, and can we get a full export before you delete our data?

Staff access is logged, and the DPA promises customers quarterly logs of all staff access, including user ID, time, action, and location. Nothing published says staff need your permission to access your data or that you can revoke that access.

• “InQuery will provide Customer with quarterly access logs showing all personnel access to Customer Personal Data” Data Processing Addendum

Ask the vendor: Do your staff need our explicit permission before viewing our case files, and can we revoke that access at any time?

InQuery mentions audit logging of file access. Nothing published describes a log of AI requests and outputs tied to users, or a way for customers to export one.

• “access controls and audit logging on every file” Plaintiff Firms page

Ask the vendor: Can we export a log showing every AI request and output on our cases, including which user ran it?

Outputs are described as first drafts and deliverables your team receives and exports, and InQuery offers an optional human QA review by its own staff. Nothing published says the tool sends anything to clients, carriers, or courts by itself. No explicit approval step or way to stop a job is published.

• “Generate a first draft of the demand based on the structured evidence already contained in the case.” Plaintiff Firms page
• “a human QA layer is available on final deliverables” Plaintiff Firms page

Ask the vendor: Can the system ever send or file anything without a person at our firm approving it, and how do we stop a job in progress?

Juro names its AI models on its own pages. The Trust Center says its AI tools are built on OpenAI's GPT model running on Microsoft Azure, and the AI Review guide lists Gemini 2.5 Pro on Google Cloud plus Claude Sonnet 4.5, Claude Haiku 4.5 and GPT-OSS on Amazon Bedrock. The DPA promises at least 30 days' notice before a new subprocessor starts handling your data, and the AI Review guide says Juro will keep customers informed of material changes such as new technology partners. Be aware the guide also says model choice may change over time, and swapping models within an existing provider would not trigger the 30-day notice. The MSA lets Juro change features at any time without a notice period.

• “Juro’s AI tools are built using OpenAI’s GPT model, provided by Microsoft on Azure servers.” Using AI responsibly with Juro (Trust Center)
• “Claude Sonnet 4.5 (hosted on Amazon Bedrock): contract analysis and issue identification.” AI Review: Data Protection and Responsible Use Guide (Help Center)
• “Juro will, at least 30 days before the New Subprocessor starts processing any Customer Data, notify Customer of the engagement” Data Processing Agreement, paragraph 10.4(a)

Juro publishes Evaluation Terms for a hands-on trial of the real platform, including the AI features (the AI Addendum is part of the trial contract), for a nominal $1. The trial ends automatically and does not roll into a paid plan. Its length is set by Juro when you start, and the website does not advertise the trial; you reach it through the sales team. Watch out: the trial is for testing only, not live work, and Juro may delete trial data at any time with no duty to help you get it back, so do not load real client files.

• “Juro will charge Customer $1 for use of the Evaluation Services in accordance with the Agreement” US Evaluation Terms (last updated 15 January 2026), clause 2.1
• “The Agreement automatically terminates on expiry of the Evaluation Period.” US Evaluation Terms, clause 3.1

Both halves are covered. The MSA bars Juro from training AI models on your data, and the MSA defines your data to include AI output. The help center says customer content is not used to train any third-party model and that Juro's AI providers have agreed not to train on your inputs and outputs. AI Extract runs through a zero-retention connection. Something to be aware of: the contract leaves room for a 'Customer-Specific Model' trained on your data and used only for your firm, which must be erased within 60 days after the contract ends. Juro's Trust Center page puts it more broadly ('we don't use your contract data to train any model'). Nothing published allows your data to train a model that serves other customers.

• “We won’t use Customer Data to train any AI models, unless the model is used only to provide Services to you” Master Services Agreement (US), Section 6.3
• “Customer content is used only to generate the requested output, and is not used to train or fine-tune any third-party AI models.” AI Review: Data Protection and Responsible Use Guide (Help Center)
• “they have agreed not to train their models using that data.” AI Addendum FAQs (Help Center)

Juro gives specific deadlines. After the contract ends there is a 60-day window to ask for your data back, and then Juro must erase the rest. The DPA says the same thing, with an exception only where the law requires Juro to keep data. AI Extract requests and outputs are deleted within 30 days unless a policy violation is found. A certificate of deletion is offered only under the EU Standard Contractual Clauses, on written request. Be aware the MSA's separate confidentiality clause lets either side keep copies in routine backups, and nothing published says when backups (taken every two hours) are purged.

• “After the 60-day period (or once we have provided the backup, if later), we must erase the remaining Customer Data in our possession.” Master Services Agreement (US), Section 2.6
• “we'll delete your AI Extract requests and outputs from our systems within 30 days.” AI Extract Data Protection Guide (Help Center)

Juro states it holds a SOC 2 Type II attestation, and the DPA commits it to keep a SOC 2 Type II report based on an audit at least every 12 months. Under the published assumptions, that claim is treated as current and covering the product. No audit period dates are published. The DPA is posted publicly on Juro's website, so you can read it before signing. A contracts tool does not need a BAA. Juro also holds Cyber Essentials, a UK government-backed certification.

• “Juro maintains a SOC 2 Type II attestation to provide comfort to its customers about the security safeguards Juro has in place.” Application security (Trust Center)
• “a SOC 2 Type II report produced by Juro’s Third-Party Auditor and updated annually based on an audit performed at least once every 12 months” Data Processing Agreement, paragraph 6.1(a)
• “This data processing agreement and its appendices ("DPA") are part of the Agreement between Juro and Customer.” Data Processing Agreement (last updated 30 September 2026)

All three parts are published. The Privacy Policy lists Juro's platform subprocessors with a place of processing for each (mostly the EEA, with Intercom and Twilio in the US). The DPA promises 30 days' notice before a new subprocessor starts processing your data, and gives you 14 days to object. Customer data is hosted on AWS in Ireland. For a US firm, that means your data sits in Europe by default. The list names AWS for hosting but does not separately mention Amazon Bedrock, which the AI Review guide says runs some of its models.

• “These are the third-party service providers who help us to process personal data on your behalf when you are a Juro customer.” Privacy Policy, section 1: Juro platform sub-processors (last updated 29 September 2026)
• “Juro will, at least 30 days before the New Subprocessor starts processing any Customer Data, notify Customer of the engagement” Data Processing Agreement, paragraph 10.4(a)
• “Customer data in Juro is hosted on Amazon Web Services (AWS) servers located in Ireland.” Application security (Trust Center)

The DPA's security appendix says Juro keeps each customer's data logically separated in its shared (multi-tenant) system. Two things back that up: the SOC 2 Type II attestation, treated as current under the published assumptions, and regular penetration tests by independent security experts. The pen test reports themselves are not published.

• “Juro also logically isolates Customer Data.” Data Processing Agreement, Appendix 2 (Security Measures)
• “Juro maintains a SOC 2 Type II attestation to provide comfort to its customers about the security safeguards Juro has in place.” Application security (Trust Center)
• “We regularly engage independent security experts to conduct penetration testing of the Juro application to identify vulnerabilities.” Network and infrastructure security (Trust Center)

Addressed, but falls short of the standard

The DPA and Trust Center promise breach notice 'without undue delay' but give no hour or day limit. The application security page says Juro notifies affected customers where required by law. The DPA says the notice will include a contact point, and Juro collects a privacy contact from your firm at signing. The general contact channel is support@juro.com.

• “Juro will notify Customer without undue delay after becoming aware of a Data Incident” Data Processing Agreement, paragraph 6.2(a)
• “Juro will notify you without undue delay (as required under Article 33(2) GDPR).” GDPR compliance (Trust Center)

Ask the vendor: Will you commit in writing to notify us of any breach affecting our data within 72 hours of discovering it, and who is our named contact?

You can get your data back in usable formats. Within 60 days after the contract ends you can ask for a copy of your latest backup, delivered within 30 days as PDF and CSV files. During the contract an admin can bulk export contracts as PDFs, but Juro switches this on only for a limited time after you ask support. After the 60-day window Juro must erase the remaining data. Written proof of deletion is offered only under the EU Standard Contractual Clauses, so ask for it if you need it.

• “We must give it to you within 30 days after your request in our standard format (currently .pdf and .csv).” Master Services Agreement (US), Section 2.6
• “If you require an export containing PDF copies of all your Juro contracts for any reason, an Org Admin can action this from the frontend” Bulk exporting documents from Juro (Help Center)
• “After the 60-day period (or once we have provided the backup, if later), we must erase the remaining Customer Data in our possession.” Master Services Agreement (US), Section 2.6

Juro says staff access follows least privilege and is logged. For AI Review, staff do not see your contract data unless you send feedback, or where needed for support or legal reasons, and that access is logged and restricted. Nothing published says staff need your explicit permission in every case, that you can see the access log, or that you can revoke access.

• “Juro personnel do not access customer contract data submitted to AI Review, unless the user chooses to provide feedback on a specific AI Review suggestion” AI Review: Data Protection and Responsible Use Guide (Help Center)
• “Access to systems is logged to create an audit trail.” Data Processing Agreement, Appendix 2 (Security Measures)

Ask the vendor: Do Juro staff need our explicit permission before viewing our contracts, can we see a log of that access, and can we revoke it?

Operator, Juro's AI chat, keeps each user's conversation history, which users can search, rename and delete. AI Review suggestions are tracked on the document record. Juro keeps its own operational logs for security and audit. Nothing published describes a complete, exportable, firm-wide log of every AI prompt and answer tied to each user, and users can delete their own Operator history.

• “Operator also remembers conversations. You can search for these, rename useful ones for ease of reference, and also delete those you no longer need” Using Juro's Operator (Help Center)
• “Operational logs may be retained for security, reliability, and audit purposes in line with Juro’s standard logging practices” AI Review: Data Protection and Responsible Use Guide (Help Center)

Ask the vendor: Can our administrators export a complete log of every AI prompt and answer across all Juro AI features, showing which user made each one?

AI Review redlines are only suggestions that a person must accept or reject, and Operator cannot take actions on individual contracts. The newer contract intake agent (in beta) acts more on its own: once an admin turns it on, it sorts incoming contracts, fills in data fields with AI Extract, and routes documents that pass the playbook straight to approvers without a lawyer's review. An admin can switch the agent on, but no published page explains how to pause it quickly or lists every automatic step in one place.

• “All redlines and comments generated by AI Review are proposed only. Users must review and approve them before making any changes to a contract.” AI Review: Data Protection and Responsible Use Guide (Help Center)
• “If no issues are flagged and the document is compliant with the playbook, it skips straight to the approvers set on the template.” Contract intake and review (Help Center, beta)

Ask the vendor: Which steps can Juro's AI agents take without a person approving them, and how do we pause or switch off an agent immediately?

Not found in public documents

Justpoint's site says it uses AI but does not name any underlying model or provider. The website Terms of Use allow changes at any time, with continued use counting as consent, but there is no model disclosure to pair with it.

Ask the vendor: Which AI model or provider powers Justpoint, and will you notify us in writing before changing it?

Not found in public documents

No pricing page, free trial, sandbox, or pilot is published. The site offers only a phone number and contact form.

Ask the vendor: Can our firm test Justpoint hands-on before signing an NDA or an annual contract?

Not found in public documents

Neither the Terms of Use nor the Privacy Policy says whether customer inputs or outputs are used to train Justpoint's models or any third-party model. The relaxed rule for anonymized data does not help, because there is no training statement of any kind.

Ask the vendor: Will you confirm in writing that our identifiable data is never used to train your models or any AI provider's models?

Addressed, but falls short of the standard

Retention is described only as lasting as long as reasonably necessary. No deletion timeframe, backup schedule, or certificate of destruction is published.

• “for as long as reasonably necessary to carry out the Company's intended business purpose for such information” Justpoint Privacy Policy

Ask the vendor: How many days after our request will you delete our data, including backups, and will you certify it in writing?

Not found in public documents

No Justpoint page claims SOC 2 of any type, so there is no claim to give the benefit of the doubt to. No DPA or BAA is published or referenced either, and PI work involves medical records.

Ask the vendor: Do you hold a SOC 2 Type II report, and can you provide it, a DPA, and a BAA before we sign?

Addressed, but falls short of the standard

The Privacy Policy lists general categories of service providers and some website analytics tools. There is no named subprocessor list for the product, no notice before new ones are added, and data location is not limited to the US.

• “may store, process, and transmit information in the United States and other locations around the world” Justpoint Privacy Policy

Ask the vendor: Can you send your subprocessor list, tell us where our data is stored, and commit to notifying us before adding new subprocessors?

Not found in public documents

No public statement says customer data is kept separate between customers. There is also no SOC 2 claim, pen test, or red-team exercise to support one.

Ask the vendor: How is our firm's data kept separate from other customers' data, and what independent audit or test checks that?

Not found in public documents

Neither published document includes a breach notification commitment or timeframe.

Ask the vendor: Will you commit in writing to notifying us within 72 hours of a breach affecting our data, and who will contact us?

Not found in public documents

No published source describes exporting customer data or destroying it afterward.

Ask the vendor: If we leave, in what format do we get our data back, and will you confirm deletion in writing?

Nothing published about when Justpoint staff can access customer data, or whether that access is logged.

Ask the vendor: When can your staff see our data, do they need our permission, and is that access logged?

Nothing published about logging AI prompts and answers or exporting such a log.

Ask the vendor: Can we export a log of every AI prompt and answer, tied to the user who ran it?

Nothing published about what the AI does on its own, human approval steps, or how a customer can stop it.

Ask the vendor: Does any AI output reach clients or courts without attorney review, and how can we turn AI features off?

Addressed, but falls short of the standard

Lawmatics names OpenAI as the model provider for Merlin. However, no published document commits to telling customers before material changes to the service. The Terms of Use reserve the right to change the service without notice and waive specific notice of changes, and the Privacy Policy only says Lawmatics 'may' notify. The customer Sales Agreement is not published.

• “Merlin is built on models from OpenAI, accessed through their developer API.” Merlin AI Suite Overview
• “you waive any right to receive specific notice of each such change.” Terms of Use
• “If we make material changes to this privacy policy, we may notify you” Privacy Policy

Ask the vendor: Will our contract commit Lawmatics to notify us in advance before changing the AI model provider or making other material changes to Merlin?

Addressed, but falls short of the standard

There is no published self-serve free trial. Lawmatics offers a 'first month free' promotion after a demo, but nothing published says whether that month requires signing a paid or annual agreement first. The Terms of Use describe annual contracts for customers who signed up before May 2023 and refer to a separate unpublished Sales Agreement for current customers. Merlin pricing is listed monthly, but contract length is not stated.

• “Get your first month of Lawmatics free” First Month Free offer page
• “Purchases are made on an annual contract basis.” Terms of Use

Ask the vendor: Can our firm get hands-on access to Lawmatics with Merlin turned on before signing an NDA or an annual contract, and does the 'first month free' offer require a signed annual agreement?

Lawmatics states on an official Help Center page that it does not train or fine-tune models on customer data and that OpenAI, its model provider, does not train on data sent through the API. Both halves are covered. No anonymized or aggregated use clause is involved. Be aware that the provider half relies on OpenAI's published API policy rather than a described contract term, and the Privacy Policy (2023) and Terms of Use say nothing about AI.

• “Your data is never used to train AI models. Not by us, and not by our AI provider.” Merlin AI Suite Overview
• “Lawmatics performs no model training or fine-tuning on customer data.” Merlin AI Suite Overview

Addressed, but falls short of the standard

The Privacy Policy mentions a 90-day post-termination limit and the Merlin page mentions 35-day backup retention, but the deletion promise carries open-ended exceptions (legal requirements, fraud, troubleshooting, investigations). The Privacy Policy is written for personal information, not clearly for the firm's case data. No deletion certificate or verification is offered, and retention for AI conversation records and the Langfuse tracing service is not stated.

• “No purpose in this policy will require us keeping your personal information for longer than 90 days past the termination” Privacy Policy
• “some information may be retained in our files to prevent fraud, troubleshoot problems, assist with any investigations” Privacy Policy
• “35-day automated backup retention and point-in-time recovery on the primary application database.” Merlin AI Suite Overview

Ask the vendor: How many days after cancellation or a deletion request are our matter data, Merlin conversation records, and AI traces (including at OpenAI and Langfuse) deleted, including backups, and will you confirm deletion in writing?

Under the amended rule, Lawmatics' SOC 2 Type 2 claim is accepted as current and covering the product, so SOC 2 is no longer a reason to fail (Lawmatics' own note that Merlin was outside its last audit period is worth knowing). Q5 still fails because no Data Processing Agreement is published or listed as available before signing, and no customer Business Associate Agreement is confirmed, even though Lawmatics markets HIPAA compliance to firms handling health information such as PI files. A fresh check of the Security page and a web search on 2026-10-06 found no DPA, BAA, or Trust Center offering them. Rule update: a DPA is no longer required to pass. If a vendor has one, it should make it available on request, so a firm should ask for it before signing.

• “Lawmatics has achieved SOC 2 Type 2 and HIPAA compliance, verified by AssuranceLab.” Security page
• “Merlin was still pre-release during our most recent audit period, so the service Merlin runs on was not part of that examination” Merlin AI Suite Overview

Addressed, but falls short of the standard

Lawmatics lists the subprocessors supporting Merlin and states data is hosted in AWS us-west-1. It does not publish a full subprocessor list for the whole platform, and no document commits to notifying customers before adding a new subprocessor. The Privacy Policy only refers generally to vendors and service providers.

• “Amazon Web Services (infrastructure hosting), OpenAI (model inference), and Langfuse (LLM observability and tracing). Twilio supports SMS delivery” Merlin AI Suite Overview
• “AWS, us-west-1, single-region for all data-plane resources.” Merlin AI Suite Overview

Ask the vendor: Do you publish a complete subprocessor list for the whole Lawmatics platform, and will you notify us in advance before adding a new one?

Lawmatics describes logical separation between firms, with each Merlin request bound to one firm's session. Independent support comes from its claimed SOC 2 Type 2 (accepted as current under the amended rule) and an annual independent penetration test whose 2026 round specifically checked that one firm cannot reach another's data. Separately, within a single firm, Merlin Copilot search does not yet respect matter-level ethical walls.

• “Tenant isolation is enforced at the ORM layer by mandatory firm scoping on tenant-owned models” Merlin AI Suite Overview
• “Our 2026 test included provisioning two separate firms specifically to verify that one firm cannot reach another firm's data.” Merlin AI Suite Overview
• “Lawmatics has achieved SOC 2 Type 2 and HIPAA compliance, verified by AssuranceLab.” Security page

Not found in public documents

No published document states how quickly Lawmatics will notify customers of a data breach. The Merlin page mentions a breach register with defined notification processes but gives no time frame, and the Terms of Use and Privacy Policy are silent on breach notice.

Ask the vendor: Will our contract commit Lawmatics to notify us of a security incident affecting our data within 72 hours, and who would contact us?

Addressed, but falls short of the standard

Customers can bulk export CRM records as CSV files, though documents and files must be downloaded matter by matter. On the destruction side, the Privacy Policy only promises to 'deactivate or delete' and keeps open-ended exceptions, so there is no clear commitment to destroy the firm's data after export or any proof of destruction.

• “Your export will come as a .csv file.” Exporting Data
• “we will deactivate or delete your account and information from our active databases.” Privacy Policy

Ask the vendor: After we export our data and cancel, will Lawmatics delete all of our data, including files, AI records, and backups, within a set number of days and confirm it in writing?

Lawmatics describes restricted staff access to production systems and supervisor-approved internal access requests. Nothing published says staff need the customer's permission to view firm data or that the customer can revoke that access.

• “only a small number of our staff can reach production systems and only through a controlled path” Merlin AI Suite Overview
• “documented and supervisor-approved access requests” Merlin AI Suite Overview

Ask the vendor: Do Lawmatics staff need our firm's explicit permission to view our data, is that access logged, and can we revoke it?

Merlin actions appear in the matter activity timeline marked as Merlin, and the Activity Hub provides a per-lead audit trail for Engage (messages sent, replies received, fields updated, escalations). Nothing published confirms that every Copilot prompt and answer is logged per user or that AI logs can be exported by the customer; Copilot users can delete their own conversation history.

• “Agent actions are written to the matter activity timeline with agent attribution.” Merlin AI Suite Overview
• “a per-lead audit trail” Merlin AI Suite Overview

Ask the vendor: Can our firm export a complete log of every Merlin prompt and response, tied to the user, including Copilot conversations?

Merlin Engage can text and email leads on its own once a firm delegates, but Lawmatics publishes exactly what it can do, offers a manual mode where a person approves every message, and lets firms revoke Merlin sessions for one user or the whole firm within five minutes. Note that initial outreach and replies are sent outside the configured outreach window.

• “Merlin Engage can run in manual mode, where every proposed message pauses for a human to approve, edit, or reject” Merlin AI Suite Overview
• “confined to sending a message, saving collected information to the matter, scheduling a follow-up, re-running scoring, and notifying your team” Merlin AI Suite Overview
• “can be revoked at any time, for a single user or for an entire firm. Revocation takes effect within five minutes.” Merlin AI Suite Overview

Not found in public documents

No official LawPro.ai page names the AI model or provider behind the product. The Terms include a general change-notice clause, but with a 'where possible' qualifier. A third-party profile says LawPro uses 'proprietary, contextualized LLMs', which would not count even if confirmed.

• “we may modify, suspend, or discontinue any part of the Services at any time with reasonable notice, where possible” LawPro.ai Terms of Service

Ask the vendor: Which AI model and provider (for example OpenAI, Azure OpenAI, Anthropic, or Google) power LawPro.ai, and will you notify us in writing before changing it?

The vendor publishes a 7-day free trial, which you request through a sign-up form on its site. Nothing published says the trial requires an NDA or an annual contract. The Fulfillment Policy says cancellation terms are set in a signed contract for paid service.

• “Try it FREE for 7 Days” Start Free Trial page

Not found in public documents

The Privacy Policy limits LawPro.ai's own training to de-identified or aggregated data. Under the amended rule, that is acceptable and covers the vendor's half. The fail now has a different cause: nothing published says whether the underlying AI model provider may train on your data. The vendor does not even name that provider.

• “Training and quality assurance, using de-identified or aggregated data only.” LawPro.ai Privacy Policy

Ask the vendor: Which AI provider powers LawPro.ai, and do you have a written agreement with it that prevents it from training on or retaining our inputs and outputs?

Addressed, but falls short of the standard

The policies say data is deleted, destroyed, or de-identified 'using NIST-approved methods' after a retention period of at least six years. They give no deletion deadline after a request or at the end of service, no backup schedule, and no certificate of destruction. The retention and deletion policies the Terms refer to are not published.

• “PHI is retained for a minimum of six years under HIPAA unless a longer period is required by law or contract.” LawPro.ai Privacy Policy
• “Data will be handled according to our retention and deletion policies.” LawPro.ai Terms of Service

Ask the vendor: How many days after we ask, or after we cancel, will you delete our data and backups, and will you give us written confirmation?

The homepage shows a SOC 2 badge, and the Privacy Policy refers to SOC 2 criteria. Under the amended rule, this counts as a SOC 2 claim (type unstated) and we assume it is current. Q5 still fails on the contract documents. No DPA is published, and no Trust Center lists one as available on request. The product handles medical records, but the Terms mention BAAs only 'where applicable' and do not confirm one is available before signing. Rule update: a BAA is required only when the product handles HIPAA-covered health information. Most plaintiff firms are not HIPAA covered entities, so this is not counted against the vendor. A firm that acts as a business associate of a healthcare provider should still ask for a BAA.

• “Data processing is governed by our Privacy Policy and Business Associate Agreements (BAAs) where applicable.” LawPro.ai Terms of Service

Not found in public documents

Subcontractors are mentioned in general terms only. No subprocessor list, no promise of notice before new subprocessors are added, and no data storage location are published.

• “Business associates and subcontractors who perform services on our behalf under HIPAA-compliant Business Associate Agreements” LawPro.ai Privacy Policy

Ask the vendor: Please send your list of subprocessors, where our data is stored, and how much notice you give before adding a new subprocessor.

Not found in public documents

The SOC 2 badge would now count as independent support. Q7 still fails because no official source says one firm's data is isolated or logically separated from other customers' data. The Case Assistant FAQ only limits each chat to the current case's data.

• “Only data tied to the current case is accessible, ensuring privacy and relevance.” Case Assistant FAQ

Ask the vendor: How is our firm's data kept separate from other customers' data in your system, and is that control covered in your SOC 2 report?

Addressed, but falls short of the standard

Both documents promise prompt notice of a breach but give no specific number of hours or days. The published contact is privacy@lawpro.ai.

• “Notifications will be made without unreasonable delay.” LawPro.ai Privacy Policy
• “LawPro.ai will promptly notify affected customers of any breach as required under HIPAA's Breach Notification Rule.” LawPro.ai Terms of Service

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a breach affecting our data?

Addressed, but falls short of the standard

You can export chronologies as PDF or Excel, and the Privacy Policy mentions a right to a portable copy of your data. No published document commits to deleting or destroying your data after export or when you leave. The Terms only refer to unpublished retention and deletion policies.

• “Chronologies can be exported as PDF or Excel files, including all annotations and references.” TimeLineD FAQ
• “Request a copy of your data in a portable format.” LawPro.ai Privacy Policy

Ask the vendor: When we leave, will you return all our data in usable formats and then delete it, with written proof of destruction?

Staff access is described as restricted, and the Privacy Policy mentions role-based access controls and audit logging. Nothing says staff need your permission before opening your data or that you can revoke their access.

• “Access is limited to authorized personnel with a legitimate business need.” LawPro.ai Terms of Service
• “Continuous monitoring, audit logging, and automated threat detection” LawPro.ai Privacy Policy

Ask the vendor: Do your staff need our permission before opening our case files, and can we see and revoke that access?

Case Assistant chats are saved per case and tied to the user who started them. Nothing published says a firm administrator can see every user's prompts and answers or export a complete log.

• “All conversations are autosaved at the case level.” Case Assistant FAQ
• “Users can only view chats that they initiated.” Case Assistant FAQ

Ask the vendor: Can a firm administrator export a complete log of every AI prompt and answer, showing which user made each one?

The tools produce chronologies, summaries, and draft documents that users review, edit, and download themselves. Nothing published says the product sends messages or files anything on its own. However, there is no stated approval step before output is used. The HPT page presents avoiding manual review as a benefit.

• “Edit, revise, and build with total traceability.” Case Assistant page
• “HPT delivers reliable results without adding friction, manual review, or additional cost.” HPT page

Ask the vendor: Does anything LawPro.ai produces go to a client, insurer, or court without a person at our firm approving it first?

Not found in public documents

The Terms include a 10-day change-notice clause, but no official page or document names the AI model or provider behind the service.

• “Such updates will become effective ten (10) days after such notice” Terms & Conditions, Section 17

Ask the vendor: Which company's AI model (for example OpenAI, Anthropic, or Google) powers LegalClerk, and will you notify us before switching it?

A self-serve free trial with per-seat monthly pricing is published, so a firm can use the real product before committing. The trial length conflicts (7 days on pricing, 3 days in the Terms), a credit card is required, and the Terms say plans auto-renew unless cancelled 30 days before the period ends and that annual commitments apply to selected plans.

• “7-Day Free Trial” Pricing
• “Cancel anytime” Pricing (Unlimited Calls plan)
• “All Plans come with an initial free trial period of 3 days.” Terms & Conditions, Section 7.7

Not found in public documents

Neither the Terms nor the Privacy Notice says anything about AI training, whether on identifiable or anonymized data. The Terms give LegalClerk a broad license to use customer data with no training exclusion. Since neither half of the training question is answered, the new anonymized-data allowance does not help.

• “to use, copy, transmit, distribute, and store” Terms & Conditions, Section 4.4

Ask the vendor: Will you confirm in writing that our callers' identifiable data and AI outputs are never used to train your models or your AI provider's models?

Addressed, but falls short of the standard

The only deletion language is a promise to respond to requests to delete Google-connected data within a reasonable timeframe. No retention periods, backup deletion schedule, or proof of destruction is published.

• “We will respond to such requests within a reasonable timeframe” Privacy Notice

Ask the vendor: How many days after a request do you delete our data, including backups, and will you certify the deletion in writing?

Not found in public documents

LegalClerk does not claim SOC 2 anywhere on its official pages, so there is no claim to give the benefit of the doubt to. The Terms refer to a Data Processing Agreement but do not publish it or offer it on request. The product handles intake calls for personal injury firms and is marketed as HIPAA compliant, yet no Business Associate Agreement is offered.

• “By entering into this Agreement, the Parties also enter into the Data Processing Agreement” Terms & Conditions, Section 13

Ask the vendor: Do you have a SOC 2 Type II report, and can you send it along with the Data Processing Agreement and a Business Associate Agreement before we sign up?

Not found in public documents

The Privacy Notice mentions unnamed service providers, but no subprocessor list, no promise to notify before adding subprocessors, and no data storage location are published. The contracting company is in Slovakia, so data location is a real question.

• “We may share data with trusted service providers who perform services on our behalf, such as cloud storage providers” Privacy Notice

Ask the vendor: Please send your list of subprocessors, tell us which country our data is stored in, and confirm you will notify us before adding new ones.

Not found in public documents

No official source says one firm's data is kept separate from another firm's. LegalClerk also makes no SOC 2 claim and publishes no penetration test, so nothing independent supports isolation either.

Ask the vendor: How is our firm's data kept separate from other customers' data, and what independent audit or penetration test confirms it?

Not found in public documents

Neither the Terms nor the Privacy Notice states any breach notification timeline. Support contact listed is support@legalclerk.ai.

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a breach involving our data?

Not found in public documents

The Terms end access at termination but say nothing about exporting data or deleting it afterward.

• “immediately cease to have access to and use the Services” Terms & Conditions, Section 12

Ask the vendor: When we leave, how do we export our call recordings, transcripts, and intake data, and will you confirm deletion afterward?

Restricted staff access and access logging are mentioned in general terms. Nothing says staff need the firm's permission or that access can be revoked.

• “Access to data is restricted to authorized personnel only.” Privacy Notice
• “LegalClerk uses HIPAA-grade encryption, secure storage, and detailed access logging.” Feature: Legal Intake

Ask the vendor: Do your staff need our permission before viewing our data, and can we see and revoke that access?

The intake page says calls are auditable and the product provides call summaries and transcripts. Nothing confirms a complete, exportable log of every AI interaction tied to a user.

• “Every call and submission is protected and fully auditable” Feature: Legal Intake

Ask the vendor: Can we export a complete log of every AI call, message, and transcript, with dates and the user or caller involved?

The AI answers calls, books appointments, and sends follow-up calls, emails, and messages to clients on its own. Nothing published describes an approval step or a way to pause or stop it.

• “automatically sends follow-up emails, messages, or calls to clients” Feature: Outbound

Ask the vendor: Can we require approval before the AI contacts a client, and how do we immediately stop all automated calls and messages?

Legora names its model providers on its own subprocessor list (OpenAI, plus Microsoft, Google and AWS for AI models) and its Trust Center says Anthropic models run on AWS. It does not say which model powers which feature. The SMB DPA requires 30 days' posted notice before any new subprocessor, including a new model provider, is engaged. The SMB terms also let firms subscribe to written notice of terms changes.

• “OpenAI, Provision of AI models” Pre-approved Sub-processors
• “Legora is hosted on Microsoft Azure with a small footprint on AWS (for Anthropic models).” Legora Trust Center FAQ
• “Legora may not engage a new sub-processor before the 30-day period has ended.” SMB Data Processing Agreement, Clause 4.2

The published SMB terms include a free trial clause. During the trial the firm pays nothing and can cancel at any time with no fees. The clause applies where Legora offers a trial, and the website does not advertise one or state its length (Legora sets it at sign-up). Warning: the trial turns into a paid subscription automatically unless the firm cancels in writing before it ends.

• “No fees are payable by the Subscriber during the Free Trial Period.” SMB General Terms and Conditions, Clause 2.4.2
• “The Subscriber may terminate the Free Trial Period at any time, for any reason, upon written notice” SMB General Terms and Conditions, Clause 2.4.8

The SMB contract says Legora will not train AI models on customer confidential information and will not let its subcontractors (which include the AI model providers) do so. Confidential information includes Subscriber Content, which the terms define as all inputs and outputs. Both halves are met. The existing record shows no use of identifiable customer data for training, so the amended rule on anonymized data does not change this result.

• “Legora will not use Subscriber's Confidential Information to train or fine-tune AI models, nor will it allow its subcontractors to do so.” SMB General Terms and Conditions, Clause 8.3.3
• “(ii) Subscriber Content” SMB General Terms and Conditions, Clause 7.1.1 (definition of Confidential Information)
• “The models we use do not retain or learn from any input or output. All processing is ephemeral.” Legora Trust Center FAQ

Legora deletes all customer content 30 days after the contract ends, deletes on written request at any time, and will certify the deletion in writing if asked. The published documents do not say when backups (taken every 4 hours) are purged.

• “After the expiry of the thirty (30) day period referred to above ... Legora shall securely delete all Subscriber Content” SMB General Terms and Conditions, Clause 4.3.2
• “Upon the Subscriber's written request, Legora shall provide written certification confirming that such deletion has been completed.” SMB General Terms and Conditions, Clause 4.3.2

Legora's own Trust Center says it holds a SOC 2 Type II report, and a 2026 audit is underway. Under the amended rule, the report is treated as current even though its listed period ended in August 2025. Legora publishes its SMB Data Processing Agreement, so it is available before signing. Legora is a research and drafting tool, not a health-records product, so a BAA is not required for this question. Firms handling health information should still ask about the HIPAA document listed in the Trust Center.

• “We currently possess a SOC 2 Type 2 report dated August 1, 2024 to August 1, 2025” Legora AB 2025 SOC 2 Bridge Letter (Trust Center)
• “This data processing agreement ... governs the processing of Personal Data in the course of the provision of the Services” SMB Data Processing Agreement

Legora publishes a named subprocessor list with a data residency entry for each provider by region (US, EU, Asia-Pacific). The SMB DPA gives firms 30 days from posting to object, and Legora cannot use a new subprocessor before that period ends. Notice works by updating the list. The DPA also bars processing outside the EU/EEA or US without the firm's consent.

• “OpenAI, Provision of AI models, Personal data included in Subscriber Content, EU/EEA” Pre-approved Sub-processors
• “Legora shall inform the Subscriber of any new sub-processors by updating the sub-processor list and give the Subscriber the opportunity to object” SMB Data Processing Agreement, Clause 4.2
• “We currently offer three processing regions; US, EU and AP.” Legora Trust Center FAQ

Legora's published Security Measures say each customer's environment is kept logically separate at all times. Two kinds of independent support back this up: the SOC 2 Type II report Legora claims, treated as current under the amended rule, and a Spring 2026 third-party penetration test summary listed in the Trust Center. Legora also commits publicly to regular third-party penetration testing.

• “The Subscriber environments are logically separated at all times.” Security Measures, Clause 12(d)
• “We currently possess a SOC 2 Type 2 report dated August 1, 2024 to August 1, 2025” Legora AB 2025 SOC 2 Bridge Letter (Trust Center)
• “Spring 2026 Penetration Test - Executive Summary” Legora Trust Center Resources

The SMB DPA commits to breach notice within 36 hours, well inside 72 hours. The Trust Center lists security@legora.com as the security contact.

• “Legora shall inform the Subscriber without undue delay and at the latest within 36 hours from becoming aware of a Personal Data breach.” SMB Data Processing Agreement, Clause 7.1

Legora will help the firm get its content back in a standard machine-readable format during the contract or within 30 days after it ends. After that it deletes the content and, on request, certifies the deletion in writing. The terms do not name specific file types such as PDF or native files.

• “assist the Subscriber in retrieving any Subscriber Content stored in the Services (if any) in a standard machine-readable format” SMB General Terms and Conditions, Clause 4.3.2
• “Legora shall provide written certification confirming that such deletion has been completed.” SMB General Terms and Conditions, Clause 4.3.2

Legora says engineers have no standing access to customer data. Access needs the customer's written, explicit consent and is temporary. The security page says access is logged as a general design principle. Nothing public says a firm can see that log or revoke access it has granted.

• “the CTO can grant temporary access, but only if written, explicit consent is provided by the customer.” Legora Trust Center FAQ
• “no user or system is inherently trusted, access is always verified, limited, and logged.” Legora Security page

Ask the vendor: Is every instance of Legora staff access to our data logged and visible to us, and can we revoke access we have granted at any time?

The SMB terms say firm administrators may be able to see logs of how users use the service. The security page describes Data Governance tools, but presents them alongside enterprise security packages, so they may not be part of the SMB plan. An Audit Logs white paper is available only on request. Public sources do not confirm that every AI prompt and answer is logged per user, or that the firm can export that log.

• “administrators may be able to access logging and information about end users' use of the Services” SMB General Terms and Conditions, Clause 2.2.1
• “Legora's Data Governance tools give you real-time insight into who's accessing your data and when.” Legora Security page

Ask the vendor: On the SMB plan, does Legora keep a log of every AI prompt and answer tied to each user, and can our firm export it?

The terms put human review of AI output on the firm, in general terms. Legora markets an Agent that plans, executes, and delivers work, plus features such as an Email Assistant and Slack intake. White papers on these are available only on request. Public pages do not say what the Agent can do without a person's approval, or how a firm can stop it.

• “The Subscriber is responsible for evaluating the accuracy and quality of Output ... including by using human review of the Output.” SMB General Terms and Conditions, Clause 6.1.1
• “The Agent plans, executes, reviews and delivers” Small and mid-sized firms page

Ask the vendor: What actions can the Legora Agent take without a user approving them first, and how can our firm turn off or stop agent features?

Confirmed by LexAxiom in an interview: the product runs on a privately hosted Qwen model (an open-weight model family from Alibaba) hosted on Google Cloud, and LexAxiom will give at least 60 days' written notice before switching models. Not yet published on the website.

• “transmitted, encrypted in transit, to the large language model (LLM) providers identified on our Subprocessor page.” Privacy Policy v3.0, Section 4.3
• “The current list is being prepared for publication.” Subprocessors page
• Confirmed by vendor interview Privately hosted Qwen base model on Google Cloud. (LexAxiom vendor interview, 2026-10-06)
• Confirmed by vendor interview Written notice at least 60 days before switching models. (LexAxiom vendor interview, 2026-10-06)

Confirmed by LexAxiom in an interview: a free two-week trial is available with no NDA and no annual contract required. Not yet published on the website. The public Terms say trial environments may lack some security features, so ask which protections are active before uploading client material.

• “Evaluation Services are provided free of charge unless an Evaluation Order states otherwise. Either party may end an evaluation at any time.” Terms of Service, Section 4.2
• “use the Evaluation Services during the evaluation period we specify” Terms of Service, Section 4.1
• Confirmed by vendor interview Free two-week trial; no NDA or annual contract required to start. (LexAxiom vendor interview, 2026-10-06)

Both halves are covered in writing. LexAxiom says it does not train any model on customer inputs or outputs, and that its AI providers are contractually barred from training on that data. The same promise applies to trial data. No de-identified or aggregated data carve-out was found, so the amended rule does not change this result. The AI providers bound by this promise are not named publicly.

• “LexAxiom does not use Customer Data or Content to train, fine-tune, or improve artificial intelligence or machine learning (ML) models” Privacy Policy v3.0, Section 5
• “This commitment binds LexAxiom and all Subprocessors.” Privacy Policy v3.0, Section 5
• “not use Customer Data or outputs to train, fine-tune, or improve their models or services” Privacy Policy v3.0, Section 4.3
• “we require our AI provider Subprocessors to accept the same restriction” Terms of Service, Section 5.5

The deletion schedule is specific. After a deletion request or termination there is a 90-day export window, deletion from live systems within 30 days, and backups overwritten within 90 days after that. Because the firm holds its own encryption key, destroying the key also makes any leftover data unreadable. No certificate of destruction is mentioned.

• “retained for a 90-day export window and then deleted from active systems within 30 days, with encrypted backups overwritten within 90 days thereafter” Privacy Policy v3.0, Section 11(c)
• “a Customer's destruction of its key renders any residual ciphertext permanently unreadable” Privacy Policy v3.0, Section 11(c)

The SOC 2 half now passes. LexAxiom claims SOC 2 Type II on its homepage and in its Privacy Policy, and under the amended rule that claim is accepted as current. The DPA half still fails. The Privacy Policy refers only to 'any' DPA, no DPA is published, there is no Trust Center listing one, and the legal page says contract documents are provided to Customers with their subscription. Nothing public says a prospective firm can get the DPA before signing. Rule update: a DPA is no longer required to pass. If a vendor has one, it should make it available on request, so a firm should ask for it before signing.

• “AES-256 encryption, SOC 2 Type II compliance, and zero data training on your client information” Homepage
• “The Customer's agreement with LexAxiom, including the End User License Agreement (EULA) and any Data Processing Addendum (DPA), governs Customer Data.” Privacy Policy v3.0, Section 2(b)
• “provided to Customers with their subscription. To request a copy, email legal@lexaxiom.com.” Legal hub

Data location (United States) is published. Confirmed by LexAxiom in an interview: Google Cloud is the only subprocessor LexAxiom uses (integrations a firm connects itself are the firm's own choice), and customers get notice before any new subprocessor is added. The public subprocessor page is still empty, so this rests on the interview.

• “A current list of Subprocessors is maintained at https://www.lexaxiom.com/legal/subprocessors” Privacy Policy v3.0, Section 6.3
• “The current list is being prepared for publication.” Subprocessors page
• “We update the list within ten (10) business days of any material change.” Privacy Policy v3.0, Section 6.3
• “The Services are hosted in the United States and in any Data Residency Region a Customer selects under the EULA.” Privacy Policy v3.0, Section 12
• Confirmed by vendor interview Google Cloud is the only subprocessor. (LexAxiom vendor interview, 2026-10-06)
• Confirmed by vendor interview Customers get notice before a new subprocessor is added. (LexAxiom vendor interview, 2026-10-06)

LexAxiom says one firm's data is never used to produce another firm's output, and each firm's data is encrypted with keys that firm controls. That is an isolation statement. Under the amended rule, the vendor's SOC 2 Type II claim counts as current independent support, so this now passes. No audit date, auditor, or pen test summary is public.

• “AI Features are configured so that no Customer's data is used to generate outputs for any other Customer.” Privacy Policy v3.0, Section 4.3
• “Customers generate and hold their own encryption keys in a key management service (KMS) they control” Privacy Policy v3.0, Section 10
• “annual System and Organization Controls (SOC) 2 Type II examination” Privacy Policy v3.0, Section 10

LexAxiom commits to notifying affected customers within 48 hours of learning about a security incident, which is shorter than the 72-hour benchmark. The contact channel for suspected account compromise is security@lexaxiom.com. 'Security Incident' is defined in the unpublished EULA, so the exact scope cannot be checked.

• “notice to affected Customers within 48 hours after we become aware of a Security Incident” Privacy Policy v3.0, Section 10(h)

Deletion at exit is published. Confirmed by LexAxiom in an interview: firms can export all of their data as CSV files, and LexAxiom will confirm in writing that the data is destroyed after the firm leaves.

• “Customer Data is retained for a 90-day export window and then deleted from active systems within 30 days” Privacy Policy v3.0, Section 11(c)
• Confirmed by vendor interview Full data export in CSV format. (LexAxiom vendor interview, 2026-10-06)
• Confirmed by vendor interview Written confirmation that data is destroyed after a firm leaves. (LexAxiom vendor interview, 2026-10-06)

Published: staff access needs approval from the firm's designated attorney-owner. Confirmed by LexAxiom in an interview: staff can see firm data only when invited by a firm member, sessions are logged, and the firm can revoke access while a session is running.

• “requires authorization by the Customer's designated attorney-owner, enforces that authorization as a technical control in the key management layer” Privacy Policy v3.0, Section 4.2
• “prohibits snapshots, exports, or recordings during any support session” Privacy Policy v3.0, Section 4.2
• Confirmed by vendor interview Staff access only by firm invitation; sessions logged; revocable mid-session. (LexAxiom vendor interview, 2026-10-06)

Confirmed by LexAxiom in an interview: an AI activity log records every AI answer, and any client data in it stays inside the firm's own tenant. Not yet confirmed: whether the log also captures every prompt and whether the firm can export it.

• “error reports and usage events are scrubbed of document text, prompts, and outputs.” Privacy Policy v3.0, Section 3.2
• Confirmed by vendor interview Every AI answer is recorded and stays in the firm's tenant. (LexAxiom vendor interview, 2026-10-06)

Ask the vendor: Does the AI activity log capture every prompt as well as every answer, tied to the user, and can we export it?

Confirmed by LexAxiom in an interview: automatic actions are off by default. The firm decides what the AI may do on its own, and the firm can turn it off at any time. The homepage language ('Ready to sign', 'No manual handoffs') describes what is possible once a firm turns automation on.

• “Every inbound call is answered, qualified, and scheduled” Homepage
• “Consultations are booked, confirmed, and reminded automatically.” Homepage
• “Outputs may be incomplete, inaccurate, out of date, or wrong” Terms of Service, Section 5.3
• Confirmed by vendor interview Automation is off by default; the firm chooses what to allow. (LexAxiom vendor interview, 2026-10-06)
• Confirmed by vendor interview The firm can turn the AI off at any time. (LexAxiom vendor interview, 2026-10-06)

Addressed, but falls short of the standard

LexisNexis openly names its model providers (OpenAI, Google, Anthropic, plus Microsoft on the subprocessor list). The notice half falls short: the US General Terms say the service and terms can change without notice. The only advance-notice promise is the DPA's 14-day notice for new subprocessors. That would cover adding a new AI provider, but not switching models within an existing provider.

• “Protégé General AI includes access to large language models from OpenAI, Google, and Anthropic.” Lexis+ with Protégé product page FAQ
• “may be changed by LN without notice to you upon posting on this site” LexisNexis General Terms and Conditions, Section 8.1
• “We will inform you of any changes by updating the list on our website at least 14 days in advance.” LexisNexis Data Processing Addendum, Section 3.1

Ask the vendor: Will you commit in writing to notify us before you change the AI model or provider behind Lexis+ with Protégé, including model changes within an existing provider?

The US product page offers a free 2-day trial of Lexis+ with Protégé, requested through a web form. Nothing on the page says you must sign an NDA or an annual contract first. Two days is short for a real evaluation.

• “Get your 2-day free trial” Lexis+ with Protégé product page

The contract says LexisNexis will not use your inputs to train large language models. The product page says customer data is never used to train AI models and is not used to train public AI models, which covers the outside model providers (OpenAI, Google, Anthropic). The Trust Center keeps a right to use aggregated and de-identified information to improve its AI services. Under the amended rule, that de-identified use is acceptable. Be aware the documents do not explain how de-identification is done.

• “will not be used by LN to train any large language model(s)” LexisNexis General Terms and Conditions, Section 4.3(b)
• “Customer data is not used to train public AI models” Lexis+ with Protégé product page FAQ
• “We never use customer data to train AI models” Lexis+ with Protégé product page, Privacy & Security
• “We may use aggregated and de-identified information ... to develop, evaluate, improve, secure, and operate our AI Services.” Trust Center, AI Training Data and Bias

The vendor gives specific retention periods. Documents uploaded in a conversation without saving to a Vault are purged at the end of the session, and conversation results are kept for 90 days. Stored files may be deleted no sooner than 60 days after the contract ends. Backup deletion schedules and a certificate of destruction are not described.

• “Uploads up to 10 documents, not saved to a Vault, are automatically purged at the end of each session.” Lexis+ with Protégé product page FAQ
• “Results from Protégé Vault are retained for 90 days under "My Conversations" history.” Lexis+ with Protégé product page FAQ
• “After a reasonable period of not less than 60 days from the effective date of termination, LN reserves the right to permanently delete all Subscriber Files” LexisNexis General Terms and Conditions, Section 4.2(b)

LexisNexis says it has a SOC 2 Type 2 report covering Lexis+ with Protégé, available on request through its Trust Center. Under the amended rule, this is treated as current. A DPA is published, and a Business Associate Addendum is published for firms handling health information. ISO 27001 is also listed.

• “Our trusted solutions, including Lexis, Lexis+ AI, Lexis+ with Protege and CaseMap+, continue to meet industry and regulatory standards.” Trust Center update, May 4, 2026
• “This report, which focuses on security, availability, and confidentiality service controls, is now available for secure download.” Trust Center update, May 4, 2026 (SOC2 Type 2 Report)
• “This LexisNexis Data Processing Addendum ("DPA") forms part of the agreement” LexisNexis Data Processing Addendum
• “for the purposes of complying with the privacy, security, and breach notification provisions of the Health Insurance Portability and Accountability Act” LexisNexis Business Associate Addendum

All three pieces are published. A product-specific subprocessor list for Lexis+ with Protégé names AWS, Anthropic, Microsoft, OpenAI, and Google, with a processing location for each. The DPA promises 14 days' notice before subprocessors change, with a right to object. Caveat: AWS locations listed include several countries besides the USA, and the list says Google's processing location can vary.

• “Lexis+ with Protégé (file uploads) ... Anthropic PBC Host Provider/Data Center/AI Functionality USA” LexisNexis Subprocessors list, updated September 17, 2026
• “We will inform you of any changes by updating the list on our website at least 14 days in advance.” LexisNexis Data Processing Addendum, Section 3.1

Not found in public documents

The SOC 2 Type 2 report would count as independent support under the amended rule, but the first requirement is still missing. No official source found says one customer's data is logically separated or isolated from other customers' data. The published statements cover related topics only: your data does not improve results for other customers, development systems are kept apart from production, and cloud workloads are isolated across environments. A fresh search on 2026-10-06 found no customer-to-customer isolation statement.

• “Customer data is not used to enhance AI performance for other customers.” Lexis+ with Protégé product page FAQ
• “We compartmentalize different environments (development, testing, staging, and production)” Trust Center, Infrastructure (Environment Segregation)
• “We isolate cloud workloads across dedicated environments to minimize risk” Trust Center, Network/Infrastructure security

Ask the vendor: Can you point us to a written statement, covered by your SOC 2 Type 2 audit, that each customer's prompts, documents, and Vault data are logically isolated from other customers' data?

Addressed, but falls short of the standard

The DPA and Trust Center promise breach notice without undue delay but give no number of hours. The Business Associate Addendum allows up to 15 business days for health-information breaches, far longer than 72 hours. The contacts listed are security@lexisnexis.com and notices@lexisnexis.com.

• “we will notify you without undue delay after becoming aware of a personal data breach” LexisNexis Data Processing Addendum, Section 7.1
• “within fifteen (15) business days after that discovery” LexisNexis Business Associate Addendum

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering any security incident affecting our data?

Addressed, but falls short of the standard

The terms let you export or retrieve uploaded files at termination, but the format is left to be agreed later. Deletion is described as a right LexisNexis reserves after 60 days, not a commitment to delete or to confirm deletion. The DPA promises to delete or return personal data at your choice at the end of service. Conversation history export is not addressed.

• “LN will make available to Subscriber, in a mutually agreed upon format and medium, a file of Subscriber Files.” LexisNexis General Terms and Conditions, Section 4.2(b)
• “LN reserves the right to permanently delete all Subscriber Files stored in the Online Services” LexisNexis General Terms and Conditions, Section 4.2(b)

Ask the vendor: At termination, will you return all our files and AI conversation history in a standard format such as PDF or native files, then delete everything and confirm the deletion in writing?

Staff access is described as role-based, least-privilege, and logged, and the terms limit LexisNexis access to saved research folders. Nothing says LexisNexis asks your permission first or that you can revoke staff access.

• “All relevant activities are logged and continuously monitored to provide robust audit trails” Trust Center, Data Access
• “Access to data is granted based on clearly defined roles and responsibilities” Trust Center, Data Access
• “LN will not access or review the content of Folders, except to the extent necessary to facilitate functions of the Online Services” LexisNexis General Terms and Conditions, Folders

Ask the vendor: Do your staff need our explicit permission before viewing our prompts or documents, and can we see and revoke that access?

Users get a 90-day conversation history of their AI work. The US security overview in the Trust Center reportedly also describes audit trails in Workrooms. Nothing published confirms a firm-level log of every AI prompt and answer, tied to each user, that the customer can export.

• “Results from Protégé Vault are retained for 90 days under "My Conversations" history.” Lexis+ with Protégé product page FAQ

Ask the vendor: Can our firm administrator export a complete log of every AI prompt and response, by user, for our records?

Protégé is marketed as an agentic assistant that runs multi-step drafting and research workflows. Output is returned to the user as drafts for review, and the terms make users responsible for verifying AI output. The US security overview in the Trust Center reportedly describes human-in-the-loop steps and admin switches to disable some features. The vendor does not publish a list of what agents can do without approval or a way to stop a running workflow.

• “Authorized Users are responsible for verifying any information provided in the Online Services.” LexisNexis General Terms and Conditions, Section 4.3(a)
• “produce a review-ready draft grounded in authoritative legal sources” Lexis+ with Protégé product page

Ask the vendor: Which Protégé agent actions can run without a person approving them, and how can our administrator stop a running workflow?

Luminance's published subprocessor list names the AI model providers it uses: OpenAI (via API, processed in the United States), plus Microsoft Azure and Google Cloud for in-cloud generative model processing. The Master Agreement requires prior written notice before amending terms, at least 12 months' notice before discontinuing a product or material functionality (with exceptions), and the DPA requires 30 days' notice before adding a subprocessor, which would include a new model provider. The documents do not say which specific model handles which feature.

• “OpenAI, L.L.C | Additional Generative Model processing, accessed via the provider's API.” Luminance Subprocessors Page
• “Luminance provides prior written notice of such proposed amendment to Customer (via email to Customer's contact)” Luminance Master Agreement, Clause 17.14

Luminance publishes Proof of Value terms for a free hands-on trial of the real product, two weeks by default, which Luminance can change in writing. The trial ends when the period expires or when a paid order form is signed, and either side can end it on seven days' notice, so it does not commit the firm to paid seats. Confidentiality terms are built into the trial terms; no separate NDA is described. You likely have to go through sales to get one.

• “The Customer may use the Product free of charge from the Commencement Date and for the duration of the PoV Period” Luminance Proof of Value Terms

Not found in public documents

Luminance's own half passes. The Trust Centre says customer data is never used to train its shared AI models. Its pseudonymised 'Usage Insights' (confidential information and identifiable data excluded) is acceptable under the amended rule. The other half is still missing. Prompts and documents used with features like Lumi go to outside model providers, including OpenAI's API, Azure and Google Cloud. No published Luminance document says those providers cannot train on or keep that data. The DPA, subprocessor list, AI Systems Terms, security page and TOMs are all silent on this. OpenAI's own public policies are not a Luminance commitment.

• “No, customer data is never used to train shared AI models.” Luminance Trust Centre FAQ
• “transmit the relevant inputs (including any query text, document content or requested redraft)” Luminance Master Agreement, Clause 10.5
• “pseudonymised data obtained from the analysis and results of Customer interactions, tagging and/or usage within the Product” Luminance Glossary, definition of Usage Insights

Ask the vendor: Can you point me to a written commitment, binding on OpenAI, Azure and Google, that they cannot train on or keep our prompts and documents?

Luminance sets specific deletion timelines. After termination there is a 30-day retention period. Data is removed from live systems within 30 days and from backups within another 30 days. Expedited deletion is available on written request. No published document offers a certificate of destruction.

• “data will be removed from data stores within 30 days and from backup data stores within a further 30 days” Luminance Technical and Organisational Measures
• “thirty (30) days commencing with the end of the Term” Luminance Glossary, definition of Retention Period

Luminance says it has regular SOC 2 Type II audits, and the report is listed in its Trust Centre (available on request). Under the amended rule we treat that audit as current and covering the product, even though no audit period is published. The DPA is published and can be read before signing. Luminance also holds ISO 27001. A BAA is not required for a contracts product.

• “Luminance undergoes regular SOC 2 Type II examinations to provide ongoing independent assurance” Luminance Security page
• “SOC 2 Type II report ... Request access” Luminance Trust Centre Resources
• “a copy of its current ISO 27001 certification” Luminance Data Processing Addendum

All three are published. The subprocessor list names each provider, its purpose and its location. The DPA promises at least 30 days' written notice before a subprocessor is added or replaced, and you can object. Your data is hosted in the AWS location on your order form, and the Trust Centre says customers may choose the AWS Region. Note that OpenAI API processing is listed as taking place in the United States, and the DPA acknowledges some processing outside the EEA/UK.

• “if we add or replace a subprocessor we follow the notice and objection process in the DPA.” Luminance Subprocessors Page
• “at least thirty (30) days' prior written notice of any intended addition or replacement of a subprocessor” Luminance Data Processing Addendum
• “Customer Data will be hosted in the AWS hosting location specified in the Product Order Form” Luminance Data Processing Addendum

Luminance says each customer gets its own isolated environment within its cloud. Independent support comes from two places: a CREST-certified outside firm runs penetration tests at least once a year, and Luminance claims SOC 2 Type II audits. Under the amended rule, that SOC 2 claim counts as current.

• “isolated environment for each customer within the Luminance cloud infrastructure” Luminance Technical and Organisational Measures
• “We conduct third-party penetration tests at least annually by a CREST-certified third-party penetration tester.” Luminance Trust Centre FAQ
• “Luminance undergoes regular SOC 2 Type II examinations to provide ongoing independent assurance” Luminance Security page

The DPA commits Luminance to telling you about a personal data breach within 24 hours of becoming aware of it, well inside the 72-hour benchmark. The Trust Centre has an 'Ask a question' channel, but no specific breach contact is published. The commitment covers personal data breaches as defined in the DPA.

• “it will inform the Data Controller within twenty-four (24) hours” Luminance Data Processing Addendum

Addressed, but falls short of the standard

The deletion half is covered. Luminance will delete or destroy all copies of your data after the 30-day retention period, and the DPA also requires deletion or return of personal data. The export half falls short. You can download your data and AI outputs during those 30 days, but no published document says what format they come in (original files, CSV, Excel or PDF). That matters most for Luminance's own work product, such as extracted clause data and review results.

• “Customer may, during the Retention Period, download and delete any Customer Data or Outputs from the Product.” Luminance Master Agreement, Clause 12.4.2
• “Luminance will delete or destroy all copies of Customer Data (including Outputs)” Luminance Master Agreement, Clause 12.4.2

Ask the vendor: In what file formats can we export our uploaded contracts, extracted clause data and AI outputs when we leave, and will you give us written confirmation of deletion?

Staff access is restricted. Staff may access your account only to provide support, under a least-privilege access policy, and that access may be made subject to your own security policies if you request it. No published document says staff need your explicit permission each time, that their access is logged and visible to you, or that you can revoke it.

• “Luminance's access to Customer's Product account is limited solely to the provision of Support.” Luminance Master Agreement, Clause 4.3
• “a formal Access Control policy governing access on the basis of least-privilege, need-to-know and need-to-use” Luminance Technical and Organisational Measures

Ask the vendor: Do your staff need our explicit permission before accessing our account, is that access logged where we can see it, and can we revoke it?

Luminance mentions audit logging in the product. Its AI Systems Terms say Luminance keeps the logs the AI system automatically generates for the life of the contract. No published document confirms that every AI prompt and answer is logged, tied to a named user, or exportable by the customer.

• “configurable role-based access, multi-factor authentication, single sign-on and audit logging.” Luminance Security page
• “Luminance shall keep the logs automatically generated by the AI System” Luminance AI Systems Terms, Article 5.3

Ask the vendor: Can we export a log of every Lumi prompt and answer, showing which user made each one?

Luminance's terms commit, in general language, to designing its AI so people can oversee it. The AI Systems Terms also refer to agentic workflows, including Lumi. Third-party coverage describes an 'Autopilot' feature that can redraft contracts and send them to the other side. Luminance does not publish a list of what the AI can do without approval, or how a customer turns off autonomous actions.

• “designed and developed in such a way...that it can be effectively overseen by natural persons” Luminance AI Systems Terms, Article 7.1

Ask the vendor: Which Lumi or Autopilot actions can happen without a person approving them, such as emailing a counterparty, and how do we require approval or switch those actions off?

MyCase's own help center names OpenAI as the model provider for 8am IQ, and the Terms of Service commit to notifying customers of material changes. Note the Terms also say changes 8am considers not material may not be individually notified.

• “The selected document is then securely sent to OpenAI, a Large Language Model (LLM)” 8am IQ FAQs
• “we will notify you as described in Section 27 of Part I” 8am Terms of Service
• “Customers notified of critical system changes that may affect their processing” About MyCase, Information for AI Assistants

MyCase offers a 10-day free trial with no credit card and publishes month-to-month pricing, so a firm can use the real product before committing. Pick the Pro or Advanced plan for the trial if you want to test the IQ features.

• “MyCase offers a 10 day no-cost trial of our legal practice management solution. There is no credit card required to start.” MyCase Pricing
• “No long-term contracts required” About MyCase, Information for AI Assistants

Both halves are covered. The Terms say 8am does not use your content to train any AI model, whether its own or a third party's, and the help center says the AI provider's terms forbid training on customer data. Things to be aware of: 8am may use inputs and outputs, through aggregated human and automated review, to monitor and improve 8am IQ. The Terms say this is not training. 8am also keeps a license to aggregated, anonymized content, and the AI provider may keep IQ content for up to 30 days. The amended rule accepts anonymized or aggregated use.

• “We do not use Customer Content or IQ Content to train or fine-tune any large language model or other generative AI model” 8am Terms of Service, 8am IQ section
• “whether developed by us or a third party.” 8am Terms of Service, 8am IQ section
• “Based on our terms with third party providers, our customers' data will not be used to train any third-party models.” 8am IQ FAQs
• “These activities do not involve training or fine-tuning any generative AI model.” 8am Terms of Service, 8am IQ section

Vendor's own documents contradict each other

The MyCase security whitepaper says uploaded documents are backed up forever, even after deletion. MyCase's own site says customer data is purged from production when a customer leaves, but gives no timeframe and says nothing about backups. No document names a specific deletion window, backup purge schedule, or certificate of destruction.

• “we store and backup any document that is uploaded to MyCase forever, even if it is deleted.” MyCase Security Whitepaper (linked from 8am Trust Center)
• “Customer data purged from production environments when customers leave the service” About MyCase, Information for AI Assistants
• “for as long as you use our Services or as necessary to fulfill the purposes for which it was collected” 8am Privacy Policy

Ask the vendor: Do you still back up uploaded documents forever, even after deletion, and if not, within how many days are our data and backups deleted after we cancel, and will you certify it in writing?

SOC 2 now passes: MyCase says it has a SOC 2 Type 2 report, and under the amended rule we assume it is current. The DPA half still fails. The Terms, Privacy Policy and MyCase site do not publish a Data Processing Agreement. MyCase's own list of documents you can request from its Trust Center has the SOC 2 report, PCI attestations, a CASA assessment, a security questionnaire and an insurance certificate, but no DPA. No BAA is published either, and the Terms bar protected health information from 8am IQ. MyCase is general practice management software, so the BAA rule for health products was not what decided this. Rule update: a DPA is no longer required to pass. If a vendor has one, it should make it available on request, so a firm should ask for it before signing.

• “SOC 2 Type 2 attested (current report available via trust.8am.com)” About MyCase, Information for AI Assistants
• “SOC 2 Type 2 report, PCI-DSS Attestations of Compliance, CASA assessment, security questionnaire, and certificate of insurance” About MyCase, Information for AI Assistants

Not found in public documents

No subprocessor list is published, and no legal document commits to a notice period before adding one. Hosting on Amazon Web Services is disclosed, but no document clearly states the country or region where data is stored.

• “Service providers who help us administer, provide, maintain, develop, improve, advertise, and market the Platform” 8am Privacy Policy
• “MyCase runs on Amazon Web Services' EC2 cloud computing platform.” MyCase Security Whitepaper

Ask the vendor: Please send your current subprocessor list for MyCase and 8am IQ, how much notice you give before adding one, and the country where our data is stored.

Addressed, but falls short of the standard

The independent support half now passes: MyCase claims SOC 2 Type 2, which the amended rule treats as current, and it cites annual outside penetration tests. The isolation statement half still falls short. The help center says safeguards keep your data from being exposed to other MyCase customers, but no official source says each firm's data is isolated or logically separated, or explains how. The whitepaper's isolation line is about other Amazon customers, not other MyCase firms. This is a close call. A written statement of per-firm separation would probably make it a Pass.

• “The 8am MyCase platform has extensive safeguards in place so that your confidential data is not exposed to another 8am MyCase customer.” 8am IQ FAQs
• “our network traffic is isolated from other AWS customers via Virtual Private Cloud (VPC) dedicated to MyCase customers only.” MyCase Security Whitepaper
• “SOC 2 Type 2 attested (current report available via trust.8am.com)” About MyCase, Information for AI Assistants

Ask the vendor: Can you confirm in writing that each firm's data, including 8am IQ prompts and outputs, is logically separated from other MyCase customers' data, and that your SOC 2 report tests that control?

Addressed, but falls short of the standard

The Privacy Policy commits to breach notice only as required by law, with no number of hours. The Terms require the customer to notify 8am within 24 hours, but set no matching deadline for 8am. Breach reports go to security@8am.com.

• “If our security system is breached, we will notify you of the breach to the extent required under applicable law.” 8am Privacy Policy
• “You will notify us by emailing security@8am.com within 24 hours if you have any reason to believe that there has been a security breach” 8am Terms of Service

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a breach affecting our data, and who will contact us?

Vendor's own documents contradict each other

Export is available as CSV files, though per the help center the full backup leaves out documents and invoices, which must be downloaded separately. On destruction, MyCase says data is purged from production when a customer leaves, but the security whitepaper says documents are backed up forever, and the Terms say 8am 'may' delete your account and may retain information under its Privacy Policy.

• “Full Backup: MyCase will create a ZIP folder that contains various CSV files” How do I get my data out of MyCase
• “we store and backup any document that is uploaded to MyCase forever, even if it is deleted.” MyCase Security Whitepaper
• “we may retain any such information in accordance with our Privacy Policy” 8am Terms of Service

Ask the vendor: After we export our data and cancel, will you delete all of our data, including backups and documents, and give us written confirmation?

MyCase says production systems are restricted to authorized employees using multi-factor login, with centralized logging. Nothing says staff need your permission before viewing your data, or that you can see or revoke that access. The Terms also allow human review of IQ inputs and outputs to improve the product.

• “Production systems accessible only via encrypted connections with multi-factor authentication for authorized employees” About MyCase, Information for AI Assistants
• “including through aggregated human and automated review” 8am Terms of Service, 8am IQ section

Ask the vendor: Do your staff need our permission before viewing our data or 8am IQ prompts, and can we see a log of and revoke that access?

Case Assistant keeps a per-user chat history that cannot be deleted. Users can copy or export individual responses, but nothing says the firm can export a full log, or that Writing and Document Assistant use is logged the same way.

• “Each Authorized User has an individual chat history associated with their use of 8am IQ Case Assistant. This chat history cannot be deleted” 8am Terms of Service, 8am IQ section

Ask the vendor: Can we export a log of every 8am IQ prompt and answer across all features, tied to each user?

8am IQ writing, summary, and chat features run only when a user asks, and the output goes back to that user to use or discard before anything reaches a client or the file. Discovery Assistant runs text recognition on documents automatically to make them searchable, but does not send or file anything. The account owner can turn 8am IQ off for the whole firm.

• “MyCase displays the LLM's output to the user, allowing them to use or discard the new modified body of text.” 8am IQ FAQs
• “we recommend that you carefully review all output prior to any use or reliance on it.” 8am IQ FAQs
• “Availability of 8am IQ is controlled by a firm-wide AI setting managed by the Account Owner” 8am Terms of Service, 8am IQ section

Not found in public documents

None of NexLaw's published pages or legal documents names the AI model or company (such as OpenAI, Anthropic, or Google) that powers the product. The Privacy Policy promises 30 days' email notice of material policy changes, but without a named model this question cannot pass.

• “NexLaw has the right to modify these Terms periodically by posting the revised terms on NexLaw's website and/or providing written notice” NexLaw Terms of Service, Section 10
• “Email notifications (30 days advance notice)” NexLaw Privacy Policy, Policy Updates

Ask the vendor: Which large language model and provider power NexLaw, and will you notify us in writing before changing that model or provider?

The Terms of Service describe a free trial with full access that you can cancel without charge, and they also describe a monthly subscription option. The trial is short, and NexLaw's pages disagree on its length (3 days in the Terms, 7 days in some marketing banners). The current pricing page shows annual pricing only and puts the trial behind a 'Request Free Trial' button. If you don't cancel, the trial converts automatically to a paid plan.

• “NexLaw offers a 3-day free trial period ("Free Trial") for new users to evaluate our Services.” NexLaw Terms of Service, Section 6(A)
• “Your Free Trial begins immediately upon account registration and provides full access to the Services.” NexLaw Terms of Service, Section 6(A)
• “Start My 3-Day Free Trial No credit card required” NexLaw Enterprise page

Vendor's own documents contradict each other

Under the new rule, NexLaw's right to create aggregated and de-identified data is acceptable and no longer counts against it. The problem is the first half of the same Terms clause. It gives NexLaw a license to use your content itself, with no de-identification required, to improve its products. Because NexLaw's product is an AI tool, that is effectively a license to use identifiable client content to improve its AI. This conflicts with the Pricing page promise that matter data is never used to improve any AI model. No binding document names the AI model provider or shows a no-training or zero-retention agreement with it. The live Terms page could not be reopened (SSL error), so this relies on the verified archived wording.

• “to improve our products and services and to create aggregated and de-identified information” NexLaw Terms of Service, Section 4(B)
• “Your matter data is never used to train or improve any AI model” NexLaw Pricing
• “Client and matter data submitted to NexLaw is not used to train public models” NexLaw Trust Center

Ask the vendor: Will you amend Terms Section 4(B) so our content can be used only in de-identified or aggregated form, and can you show us the no-training commitment from your AI model provider?

Addressed, but falls short of the standard

Firms can set their own retention rules, but no document gives a deletion deadline after a request, a schedule for deleting backups, or a certificate of destruction. After cancellation, the policy says only that data 'may' be deleted.

• “Customer Data: Per customer settings and legal requirements” NexLaw Privacy Policy, Data Retention Periods
• “After service termination, data may be permanently deleted” NexLaw Refund and Cancellation Policy, Article 8.1
• “Firms can set retention rules so that matters and related AI outputs are retained or deleted” NexLaw Trust Center

Ask the vendor: How many days after our request or cancellation will you delete our data, including backups, and will you give us written confirmation?

Under the benefit-of-the-doubt rule, SOC 2 now counts. The Pricing page says NexLaw is SOC 2 Type II certified, so we assume the audit is current and treat the Trust Center's 'aligned with' and 'readiness' wording as old text. This question still fails on the BAA. NexLaw markets to personal injury and medical malpractice firms, whose files contain medical records. A BAA appears only in a blog post, and no legal document or Trust Center page confirms one is available before signing. The DPA is also unclear. None is published, and the DPA, BAA, and subprocessor URLs return not found. The Pricing page lists a DPA as a Professional-tier feature, while the pricing FAQ says one comes with every deployment. Rule update: a BAA is required only when the product handles HIPAA-covered health information. Most plaintiff firms are not HIPAA covered entities, so this is not counted against the vendor. A firm that acts as a business associate of a healthcare provider should still ask for a BAA.

• “SOC 2 Type II certified Independently audited. Full report available under NDA for enterprise reviews.” NexLaw Pricing
• “Custom Data Retention & DPA” NexLaw Pricing, Professional tier

Not found in public documents

No subprocessor list is published. The Privacy Policy names only categories of providers plus AWS and Google Cloud, and does not name the AI model provider. No document promises notice before a new subprocessor is added. Data location is also inconsistent: the pricing page says US-only with no cross-border transfers, while the Privacy Policy allows international transfers and the Enterprise page offers 'US or Global regions.'

• “Cloud hosting (AWS, Google Cloud) and infrastructure services” NexLaw Privacy Policy, Section 4
• “All data stored and processed on US servers. No cross-border data transfers.” NexLaw Pricing
• “NexLaw operates globally and may transfer your information to countries outside your location” NexLaw Privacy Policy, Section 7

Ask the vendor: Can you send your full subprocessor list (including the AI model provider), tell us how much notice you give before adding one, and confirm in writing that our data stays in the US?

The Trust Center says each firm's data is logically isolated from other customers. Under the amended rule, NexLaw's claim of SOC 2 Type II certification counts as current independent support for that statement, so this hard stop now passes.

• “Each firm has logically isolated environments and access controls to prevent cross-matter or cross-client data leakage” NexLaw Trust Center
• “SOC 2 Type II certified Independently audited.” NexLaw Pricing

Not found in public documents

No published document says how quickly NexLaw will notify customers of a data breach. The Trust Center mentions incident response procedures but gives no timeline and no notification promise. The only contact listed anywhere is info@nexlaw.ai.

• “Change management, access management, and incident response procedures formally defined and tested” NexLaw Trust Center

Ask the vendor: Will you commit in writing to notifying us within 72 hours of discovering a breach affecting our data, and who will contact us?

Addressed, but falls short of the standard

The cancellation policy puts the job of exporting data on the user but does not say which formats are available. Help after cancellation is 'limited' and may cost extra. It says data 'may' be deleted after termination but does not commit to deleting it or confirming that it was destroyed.

• “NexLaw may provide limited data export assistance for a fee after cancellation” NexLaw Refund and Cancellation Policy, Article 8.2
• “Users are responsible for exporting any data they wish to retain before service termination” NexLaw Refund and Cancellation Policy, Article 8.1

Ask the vendor: In what file formats can we export all our matters, documents, and AI outputs, and will you give written confirmation that our data was deleted after we leave?

Staff access is described as restricted. However, the Terms let staff access your content without your consent for broad reasons such as performing the service or investigating errors. No document says staff access is logged for your review or that you can revoke it.

• “with the consent of User or to the extent reasonably required to perform the Services, investigate system errors, comply with legal requirements” NexLaw Terms of Service, Section 8(B)
• “access to client data is strictly limited to authorized personnel who require it for specific tasks” NexLaw Security page

Ask the vendor: Will you require our permission before staff view our data, log every staff access, and let us revoke that access?

Audit logs record 'AI analyses' at the matter level, and administrators can export them. Nothing confirms that the log captures the full text of every prompt and every AI answer.

• “Matter-level actions such as document uploads, edits, exports, deletions, and AI analyses” NexLaw Trust Center, audit trails
• “Authorized administrators can export audit logs for internal review, regulatory reporting, or production” NexLaw Trust Center, audit trails

Ask the vendor: Does your exportable audit log capture the full text of every AI prompt and answer, tied to the user who made it?

NexLaw is described as a research, drafting, and trial-preparation tool that returns output to the attorney. Nothing published says it sends messages, files documents, or acts on its own. The documents tell users to review output, but only in general terms. NexLaw does not publish a description of approval steps or of what the AI can and cannot do without a person signing off.

• “The output generated by these machine learning models is probabilistic and should be assessed for accuracy” NexLaw Terms of Service, Section 9(B)
• “users should critically review important results and cross-check as needed” NexLaw Help Center FAQ

Ask the vendor: Can any NexLaw feature send, file, or share anything without an attorney approving it first, and if so, how do we turn that off?

Not found in public documents

NextPhone does not name the AI model or provider behind the receptionist anywhere in its official materials. The Privacy Policy lists only generic categories such as voice AI and AI language processing providers. The Terms do promise notice of material changes by app or email, but the model half of this question is missing.

• “voice AI and transcription, AI language processing” NextPhone Privacy Policy
• “We will notify you of material changes through the app or by email.” NextPhone Terms of Service

Ask the vendor: Which AI model providers (for voice, transcription, and language processing) power NextPhone, and will you notify us before switching them?

A firm can start a 7-day free trial, set the product up itself in about five minutes, and then continue month to month with no annual commitment. The vendor's facts file says no card is required; the pricing page says you are not charged if you cancel before the trial ends.

• “Month-to-month, no contracts, no cancellation fees” NextPhone Pricing
• “7-day free trial, no card, cancel anytime.” NextPhone llms-full.txt

Not found in public documents

Neither the Terms nor the Privacy Policy says whether call recordings, transcripts, or intake answers are used to train NextPhone's AI or the AI of its outside model providers. The license is limited to providing, securing, and supporting the service, and data is not sold, but there is no statement about training at all. The new allowance for anonymized data does not help here, because the problem is silence, not a de-identified use clause. Silence on either half still fails this hard stop.

• “you grant us the limited rights needed to host, process, and transmit that data to provide, secure, and support the Service.” NextPhone Terms of Service

Ask the vendor: Will you confirm in writing that our call recordings, transcripts, and intake data are never used in identifiable form to train your models or any AI provider's models, and that your providers are contractually bound to the same?

Addressed, but falls short of the standard

The Privacy Policy commits to deleting data from active systems within 30 days of a request, which is a real timeline. But backups are kept for an undefined limited period, and the vendor may keep limited records for legal, billing, or security reasons without saying how long. There is no certificate of deletion or verification offered.

• “we delete it from our active systems within 30 days of your request and instruct our service providers to do the same.” NextPhone Privacy Policy
• “Copies may persist in encrypted backups for a limited period before being purged” NextPhone Privacy Policy

Ask the vendor: How many days after a deletion request are backup copies purged, what records do you keep after that, and can you provide written confirmation of deletion?

Not found in public documents

The new benefit-of-the-doubt rule only applies when a vendor says it has SOC 2. NextPhone makes no SOC 2 claim of any kind, and no ISO or other audit, on any official page. It has no security page or Trust Center (/security and /trust return not found). No Data Processing Agreement is published or listed as requestable (/dpa returns not found). The Terms also say NextPhone does not sign BAAs and prohibit use with protected health information, which matters for personal injury intake calls that may include medical details.

• “We do not offer or execute Business Associate Agreements (BAAs).” NextPhone Terms of Service

Ask the vendor: Do you have a SOC 2 Type II report covering NextPhone, can you provide a Data Processing Agreement before we sign up, and how should we handle intake calls that include medical information given you do not sign BAAs?

Addressed, but falls short of the standard

Data location is published (United States). Service providers are described only by category, not by name, for the product itself (PostHog, Meta, and Google are named only for the public website). There is no commitment to notify customers before adding a new provider.

• “We rely on a small set of service providers to run NextPhone: telephony and text-message delivery, voice AI and transcription” NextPhone Privacy Policy
• “Data is stored in the United States.” NextPhone Privacy Policy

Ask the vendor: Can you send a named list of every subprocessor that handles our call data, and will you notify us before adding a new one?

Addressed, but falls short of the standard

NextPhone says recordings and transcripts are restricted to the firm's own users. That is an access-control statement, not a description of how customer data is separated in its systems or AI processing. Even read generously as an isolation statement, there is no SOC 2 claim (so the new benefit-of-the-doubt rule cannot apply) and no published pen test or red-team exercise to support it.

• “Recordings and transcripts are restricted to your account's users.” NextPhone llms-full.txt

Ask the vendor: How is our data separated from other customers' data, including in the AI's caller memory, and what independent audit or penetration test verifies that?

Not found in public documents

Neither the Terms nor the Privacy Policy contains any commitment on when or how customers will be notified of a data breach. Only a customer obligation to report unauthorized account access is stated.

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a breach affecting our data, and through what contact?

Addressed, but falls short of the standard

Customers can request a copy of their data and request deletion, but no document describes an export format (such as audio files, PDF, or CSV of call records and contacts) or a self-serve export tool. Deletion is described, but no proof of destruction is offered.

• “Request a copy of your data” NextPhone Privacy Policy
• “When you request deletion, we delete the data associated with your account as described in our Privacy Policy.” NextPhone Terms of Service

Ask the vendor: In what format can we export all recordings, transcripts, and contact records when we leave, and will you confirm deletion in writing afterward?

Nothing public describes whether or when NextPhone staff can access a firm's recordings or transcripts, whether that access is logged, or whether the firm must give permission.

Ask the vendor: Under what circumstances can NextPhone employees access our call recordings and transcripts, is that access logged, and do we have to approve it?

Every call gets a recording, transcript, and AI summary, which serves as a record of what the AI said to callers, and these are pushed to email, SMS, or a connected CRM. Nothing published confirms the firm can export that full log in bulk or that it records which internal user configured or changed the AI.

• “recording, transcript, AI summary, extracted intake fields, and a classified outcome for every call.” NextPhone llms-full.txt

Ask the vendor: Can we export a complete log of every call transcript and AI summary, with dates, in bulk?

The AI acts on its own during calls: it qualifies matters, books consultations, texts callers, and can send payment links, with no human approval step. The Terms place responsibility for what the AI says on the firm. Escalation transfers to a human are described, but no published way to pause or stop the AI or limit what it can say without approval.

• “Statements made by the assistant ... are made on your behalf and are your responsibility.” NextPhone Terms of Service

Ask the vendor: How can we immediately pause the AI receptionist, and can we require approval before it books consultations, sends texts, or quotes fees?

Vendor's own documents contradict each other

The subprocessor list names Google Cloud, Microsoft Azure, and Amazon Web Services as cloud and language model providers, but no specific model is named. The Terms of Service promise prior notice of scheduled changes in one place, then reserve the right to change the service without notice in Section 15, so the notice commitment is contradicted.

• “Google Cloud Platform | Cloud computing infrastructure, Language Learning Model | US” Subprocessor List
• “We will provide you with prior notice of any scheduled changes to the Services you are using.” Terms of Service
• “modify the Services at any time or for any reason without notice to you.” Terms of Service, Section 15

Ask the vendor: Which specific AI models power Paxton today, and will you commit in writing to notify us before switching models or making other material changes to the service?

A 7-day self-serve free trial and month-to-month billing are published. Note that the Terms say the account is charged for the chosen subscription when the trial ends, so cancel before day 7 if you do not want to continue.

• “We offer a 7-day free trial to new users who register with the Services.” Terms of Service, Section 6
• “We offer month-to-month plans with no long-term commitments.” Help Center FAQ

The DPA covers both halves. Paxton will not train models on identifiable customer personal data, and it must require its AI providers to make the same promise. Paxton may use customer data to improve the service only after it is de-identified, which the amended rule accepts. Be aware the blog's 'never trained on your data' language is broader than the contract, which allows de-identified use. The promise is written around personal data in your files, not every word of firm content.

• “Paxton may use Customer Personal Data to improve the Services only after such data has been converted into Deidentified Data” Data Processing Addendum, Section 2
• “Paxton will not use Customer Personal Data to train or fine-tune any generalized artificial intelligence or machine learning model.” Data Processing Addendum, Section 6
• “Paxton will require its artificial intelligence subprocessors not to use Customer Personal Data to train or fine-tune their generalized models.” Data Processing Addendum, Section 6

Addressed, but falls short of the standard

Deletion after termination follows a 30-day export window but is tied to unpublished retention policies, and the backup cycle length is not stated. No deletion deadline, fixed backup schedule, or certificate of destruction is published.

• “Paxton will delete Customer Personal Data from its active production systems in accordance with its retention policies” Data Processing Addendum, Section 10
• “Residual copies may remain in backups until deleted through Paxton's ordinary backup cycle” Data Processing Addendum, Section 10
• “We will only keep your personal information for as long as it is necessary for the purposes set out in this privacy notice” Privacy Policy

Ask the vendor: How many days after termination or a deletion request is our data deleted from production and from backups, and will you provide written confirmation of deletion?

SOC 2 now counts: the Trust Center says Paxton is SOC 2 compliant, so under the amended rule the audit is assumed current. A self-serve DPA is published. But Paxton handles medical records and markets medical chronologies, and no official source confirms a BAA is available before signing. The DPA only says PHI cannot be submitted unless a separate BAA is signed. The security and help center pages mention HIPAA, but neither offers a BAA. Rule update: a BAA is required only when the product handles HIPAA-covered health information. Most plaintiff firms are not HIPAA covered entities, so this is not counted against the vendor. A firm that acts as a business associate of a healthcare provider should still ask for a BAA.

• “SOC 2 ... Compliant” Paxton Trust Center
• “personal data contained in litigation documents, pleadings, correspondence, contracts, medical records, medical billing records, medical chronologies” Data Processing Addendum, Schedule 1
• “Customer must not submit Protected Health Information ... unless Customer and Paxton have entered into a separate written business associate agreement” Data Processing Addendum

A subprocessor list with purposes and US locations for every entry is published, and customers can sign up for notices of proposed new subprocessors. The list was last updated March 5, 2025, and no advance-notice period in days is stated.

• “Paxton provides notice via this advisory of updates to the list of sub-processors that are utilized, or which Paxton proposes to utilize” Subprocessor List
• “Microsoft Azure | Cloud computing infrastructure, Language Learning Model | US” Subprocessor List

Not found in public documents

Paxton's claimed SOC 2 would now count as independent support, but no reviewed source says customer data is isolated or logically separated between customers. That includes the security page, Trust Center, help center, DPA (Schedule 2 security measures), Terms, and Privacy Policy. On 2026-10-06 the security page and help center were checked again and still say nothing about separation.

Ask the vendor: How do you keep our firm's data separated from other customers' data, and is that separation covered by your SOC 2 audit?

Addressed, but falls short of the standard

The DPA commits to notice without unreasonable delay but gives no specific number of hours. The Terms of Service and Privacy Policy say nothing on breach notice timing.

• “Paxton will notify Customer without unreasonable delay and within any time period required by applicable Data Protection Law.” Data Processing Addendum, Section 8

Ask the vendor: Will you commit in writing to notify us of any security incident affecting our data within 72 hours, and through what contact?

Addressed, but falls short of the standard

Users can download generated documents and query results as Word files, and the DPA allows an export request for 30 days after termination. But the DPA does not say what format the full data return comes in, deletion afterward is tied to unpublished retention policies with backups kept for an unstated cycle, and no proof of destruction is offered.

• “Customer may request export of Customer Personal Data for thirty (30) days following termination or expiration of the Agreement.” Data Processing Addendum, Section 10
• “You can export generated documents and query results as .docx files” Help Center FAQ
• “Paxton will delete Customer Personal Data from its active production systems in accordance with its retention policies” Data Processing Addendum, Section 10

Ask the vendor: When we leave, in what format will you return all of our matters, uploaded files, and outputs, and will you give us written confirmation once everything, including backups, is deleted?

Staff access is restricted by role and reviewed quarterly, and production systems are logged and monitored. Nothing says staff need the customer's permission to view data, or that customers can revoke such access.

• “role-based access to Customer Personal Data, limited to personnel who need such access to perform their job functions.” Data Processing Addendum, Schedule 2
• “Paxton follows the principle of least access and conducts quarterly access reviews” Help Center: Security & Compliance

Ask the vendor: Do your staff need our explicit permission before viewing our matters or documents, and can we see and revoke that access?

No official source describes a customer-exportable log of AI prompts and answers tied to individual users. The organization settings page covers members, roles, and billing only.

Ask the vendor: Can an admin export a log of every AI prompt and response in our account, showing which user made each one?

Paxton produces research and drafts inside the app for the lawyer to review, edit, and export. Nothing found shows it sends messages or files anything on its own. No document states a formal approval requirement, and none describes what its multi-step drafting workflow can do without approval.

• “Canvas is Paxton's in-app editor that allows legal professionals to seamlessly revise and finalize AI-generated documents” Help Center index: Editing Drafts with Canvas

Ask the vendor: Can any Paxton feature send, file, or share content outside the app without a user approving it first, and how would we stop it?

The vendor's own blog names OpenAI (o1, previously GPT-4) as the model behind medical chronologies, and the current Terms promise advance notice of material adverse changes to paid subscriptions. Caveats: the model post is from 2024 and covers one feature, and the Terms say the vendor 'may use or change third-party AI models and processors.' The older Help Center Terms say it may use a different model than the one shown to the user.

• “ProPlaintiff.ai now uses OpenAI o1 technology” Blog: AI-Powered Medical Chronologies with OpenAI o1 (Sept 23, 2024)
• “For a material adverse change affecting a paid subscription, we will provide reasonable advance notice by email or in-product notice” Terms of Service, Section 14

The FAQ says firms can get a limited-time trial after a demo, and the Terms describe cancelling during a trial without being charged. The trial length isn't published, and the wording 'can receive' doesn't promise a trial to everyone. Paid plans are custom per-seat contracts billed annually.

• “When you book a demo, your firm can receive a limited-time trial of ProPlaintiff.ai” FAQ
• “Trial cancellation before the stated trial end prevents a first subscription charge” Terms of Service, Section 7

Vendor's own documents contradict each other

Under the new rule, the Terms clause allowing use of aggregated or de-identified operational information to improve the service is acceptable and no longer counts against the vendor. The FAQ statement that data is never used to improve public models like ChatGPT or Claude reasonably covers the model provider half. The remaining problem concerns identifiable data. The binding Terms only rule out training 'generalized or non-personalized' models on Customer Content, which leaves training personalized models on the firm's own identifiable content open. That conflicts with the Privacy Policy's promise that data will 'never' train AI models. A re-check of the current Terms on 2026-10-06 found no other sentence about training or personalized models.

• “We will not use Customer Content to train generalized or non-personalized AI models.” Terms of Service, Section 3
• “Your data will never be used to train AI models or sold to third parties.” Privacy Policy (Help Center)
• “never used to improve public models like ChatGPT or Claude” FAQ

Ask the vendor: Will you confirm in writing that our identifiable inputs and outputs are not used to train any model, including personalized models, and that your AI model providers are contractually barred from training on or retaining our data?

Addressed, but falls short of the standard

Deletion is described only as 'in accordance with our security protocols.' The Terms hand backup deletion timing to the Privacy Policy, but the published Privacy Policy gives no timing. The vendor's /privacy page actually shows a copy of the Terms. No deletion deadline, backup schedule, or certificate of destruction is published.

• “it is purged from our active systems in accordance with our security protocols” FAQ
• “The applicable Privacy Policy governs deletion timing for Google data and backup copies following a valid deletion request.” Terms of Service, Section 8

Ask the vendor: How many days after a deletion request are our files removed from your live systems and from backups, and will you provide written confirmation of deletion?

Under the benefit-of-the-doubt rule, the vendor's SOC 2 Type II claim is now treated as current and covering the product, so the SOC 2 part passes. The item still fails on the DPA. No DPA is published, and the Terms mention only a 'separately executed' DPA 'where applicable'. There is no Trust Center listing a DPA as requestable, and no other page says a DPA is offered before signing. The Terms do require a BAA before health information is uploaded, which suggests a BAA exists. Neither the Terms nor the HIPAA blog say a standard BAA is offered to every customer before signing. Rule update: a BAA is required only when the product handles HIPAA-covered health information. Most plaintiff firms are not HIPAA covered entities, so this is not counted against the vendor. A firm that acts as a business associate of a healthcare provider should still ask for a BAA.

• “ProPlaintiff currently publicly states that it is SOC 2 Type II compliant and HIPAA compliant.” Blog: SOC 2 Compliant Legal Software
• “the parties must execute a BAA before Customer submits protected health information in that capacity” Terms of Service, Section 5

Not found in public documents

The FAQ says data is stored in the United States. No subprocessor list is published, and nothing commits to notifying customers before a new subprocessor is added.

• “Your firm's data is stored on secure servers located within the United States.” FAQ
• “We may use subcontractors but remain responsible for their performance under these Terms.” Terms of Service, Section 14

Ask the vendor: Can you provide your current subprocessor list, including AI model providers, and commit to notifying us before adding a new one?

Not found in public documents

The vendor's SOC 2 Type II claim now counts as independent support. But the isolation test also requires the vendor to say that customer data is isolated or logically separated between firms, and no Terms, FAQ, Privacy Policy, Help Center, SOC 2 blog, or HIPAA blog page says so. Without that statement this hard stop still fails.

Ask the vendor: How is our firm's data kept separate from other customers' data, including in AI prompts and search, and where is that stated in writing?

Addressed, but falls short of the standard

Breach notice is promised only 'as required by applicable law', with no specific deadline. The contact channels listed are info@proplaintiff.ai and cx@proplaintiff.ai.

• “We will give notice of a confirmed security incident affecting Customer Content as required by applicable law and any controlling BAA or DPA.” Terms of Service, Section 5

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a security incident affecting our data?

Addressed, but falls short of the standard

Export in DOCX and PDF is described, with a 30-day export window after termination. But the vendor only says it 'may' delete or disable access afterward. That is not a commitment to destroy the data, and there's no proof of destruction.

• “You can export any generated Demand Letters, Chronologies, or original files as .DOCX or .PDF at any time.” FAQ
• “We may then delete or disable access to remaining Customer Content, subject to legal holds, a BAA or DPA, and the Privacy Policy.” Terms of Service, Section 8

Ask the vendor: After we export our data and leave, will you commit to deleting all copies, including backups, by a set date and give us written confirmation?

Staff access is described as limited to essential personnel for maintenance and security. Nothing says staff need the customer's permission, that their access is logged, or that the customer can revoke it.

• “Only essential personnel may access data for system maintenance and security.” Privacy Policy (Help Center)

Ask the vendor: Do your staff need our explicit permission before accessing our case data, and is that access logged and revocable by us?

Admins can view System Logs of user activity, setting changes, and administrative actions. The vendor doesn't confirm the logs capture every AI prompt and answer, or that they can be exported.

• “System Logs provide a complete audit trail of activity within your organization.” Help Center: Managing Your Organization

Ask the vendor: Do your System Logs record every AI prompt and response by user, and can we export them?

Automations can send emails, including to the case client, and can create tasks, generate documents, and update case information without a person approving each action. The Help Center describes choosing which tools an automation can use, testing on an existing case, and enabling an automation case by case, but it does not describe a per-action approval step or a way to stop all automations at once. The Terms mention approval steps only in general terms.

• “The Email tool allows automations to send emails automatically.” Help Center: Automations
• “Customer controls authorization and permissions for those actions, should review settings and approval steps” Terms of Service, Section 4

Ask the vendor: Can we require a person to approve every AI-drafted email or record change before it is sent or saved, and how do we stop all automations at once?

Addressed, but falls short of the standard

The model half passes: Robin's own DPA names Anthropic Claude through AWS Bedrock and lists Anthropic and Amazon Bedrock as subprocessors. The notice half falls short. The Terms promise email notice only for material changes to the Terms themselves, and separately say Robin may update the software without prior notice. No published document commits to telling customers before a change to the service or the underlying model.

• “AI models (e.g., Anthropic Claude via AWS Bedrock) process data within isolated tenant contexts.” Data Processing Agreement, Annex B Data Segregation
• “If we make material changes to these Terms, we will notify you via Services by emailing to the address associated with your account.” Terms and Conditions, clause 1.16
• “update the Software from time to time without prior notice to you.” Terms and Conditions, clause 1.7

Ask the vendor: Will you commit in writing to notify our firm before you switch or materially change the AI model or provider that processes our data?

Addressed, but falls short of the standard

The Terms describe a discretionary free trial and a subscription that continues on a monthly basis, but trials can be withdrawn without notice and the website offers no public sign-up or trial page, only a demo request form. A firm cannot confirm from public sources that it will get hands-on access.

• “We may make the Software available to you on a free trial basis for the limited period stated when you sign up.” Terms and Conditions, clause 1.8
• “Fill out your information below to meet with our team and see Robin's Legal Intelligence Platform in action.” Book a Demo page

Ask the vendor: Can our firm get a free trial or month-to-month account to test Robin hands-on before signing an NDA or annual contract, and how long is the trial?

Vendor's own documents contradict each other

The de-identified and aggregated data clause (Terms 1.9) is acceptable under the amended rule and is not the problem. The problem is Terms clause 2.2, which says Robin stores your actual inputs and outputs to train its models, and the user content licence covers training AI models. That is training on identifiable client content, and it conflicts with the Security page, which says no training without express consent. Separately, no public document says Anthropic or AWS is barred from training on your data.

• “we will process and store your inputs as well as the outputs from the platform, for the purposes of training our models.” Terms and Conditions, clause 2.2
• “Customer data will not be used for model training, fine-tuning, or other feature development without express consent.” Security at Robin
• “We may retain de-identified or aggregated data derived from your use of the Software for internal analytics or product improvement.” Terms and Conditions, clause 1.9

Ask the vendor: Your Terms say you train your models on our inputs and outputs, but your Security page says you do not without consent; which controls, and will you confirm in writing that neither Robin nor Anthropic or AWS will train on our identifiable data?

The DPA commits to deleting customer data on written request and confirming deletion within 60 business days, and states backups are kept for 30 days. Deletion on termination is qualified by 'to the extent technically possible,' and 60 business days is a long window.

• “Upon written request from the Customer, the Company will delete the Customer Data, and shall provide confirmation to the Customer of the deletion within 60 business days.” Data Processing Agreement
• “Continuous point-in-time recovery for S3 buckets and databases, with 30-day retention.” Data Processing Agreement, Annex B Data Backups

Robin's own DPA claims SOC 2 Type II, which under the amended rule is assumed current and to cover the platform even though no audit date is published and the linked Trust Center was unreachable. A DPA is published on the website, so it is available before signing. Contract review does not normally involve health records, so a BAA is not required.

• “Robin's security measures are validated through SOC 2 Type II, ISO 27001:2022, and Cyber Essentials certifications” Data Processing Agreement, Security measures
• “Robin's security systems are ISO, and SOC 2 certified.” Security at Robin

The DPA lists named subprocessors (including AWS, Anthropic, Amazon Bedrock, SageMaker, HubSpot, Snowflake, Mixpanel, Sentry), gives notice of new ones by updating the website list with a right to object, and names processing countries. Weak points: notice is only a website update, and the country list ends with 'any other jurisdiction where necessary.'

• “Anthropic: Hosted software services: API based AI Services” Data Processing Agreement, Subprocessors
• “notice to the Customer regarding the appointment of each subcontractor by updating the list of subprocessors on its website” Data Processing Agreement, clause 8.1
• “USA, UK, Singapore, South Africa, Australia, and Germany, and in any other jurisdiction where such processing is necessary” Data Processing Agreement, cross-border transfers

The DPA states customers are isolated by unique tenant IDs, including during AI processing. The vendor-claimed SOC 2 Type II now counts as current independent support under the amended rule. No pen test or red-team results are published, but none is needed for a Pass.

• “Unique tenant IDs enforce complete isolation between customers.” Data Processing Agreement, Annex B Data Access Controls
• “Robin's security measures are validated through SOC 2 Type II, ISO 27001:2022, and Cyber Essentials certifications” Data Processing Agreement, Security measures

The DPA commits to notifying the customer of a personal data breach within 72 hours. The notice address for the vendor in the DPA is compliance@robinai.com.

• “within 72 hours notify the Customer if it becomes aware of” Data Processing Agreement, Personal Data Breach

The DPA lets the customer get a copy of its data in a format the customer reasonably specifies, and commits to deleting or destroying data on termination, plus deletion with written confirmation within 60 business days on request. Both halves are present, though deletion on termination is qualified by 'to the extent technically possible.'

• “a copy of or access to all or part of any Customer Data in its possession or control in the format and on the media reasonably specified by the Customer.” Data Processing Agreement, Data return and destruction
• “the Supplier to the extent technically possible will securely delete or destroy or, if directed in writing by the Customer, return and not retain” Data Processing Agreement, Data return and destruction

Staff access requires the customer's prior authorisation for support, and access is logged. Nothing public says the customer can revoke that access.

• “Customer data is accessible only to customers unless explicit prior authorisation is provided for support requests.” Data Processing Agreement, Annex B Data Access Controls
• “All logs (authentication, access, system activities) are aggregated in AWS CloudWatch and Datadog, with immutable storage” Data Processing Agreement, Annex B

Ask the vendor: Can our firm revoke Robin staff access to our data at any time, and can we see the log of staff access?

Robin says input activity is logged, but nothing public confirms a per-user log of every AI prompt and answer that the customer can export.

• “All input activities are logged with tenant context.” Data Processing Agreement, Annex B Input Controls

Ask the vendor: Can our firm export a log of every AI prompt and answer, showing which user made it?

Robin is described as a review, search, and drafting assistant whose answers carry citations for a lawyer to check, and nothing public shows it sending messages or filing on its own. There is no published statement of approval gates or a stop mechanism for any automated features.

• “Click on citations to verify answers quickly in the underlying primary documents.” Legal Intelligence Platform

Ask the vendor: Does any Robin feature act or send anything without a user approving it first, and how can we switch such features off?

Not found in public documents

No Smith.ai page or legal document we reviewed names the AI model or provider behind the AI Receptionist. The Terms let Smith.ai change terms effective on posting; advance notice (30 days) is promised only for Free Plan pricing and call allowance changes, not for changes to the service or its AI.

• “All changes are effective immediately when we post them and apply to all access to and use of the Platform and Services thereafter.” Smith.ai Terms of Service

Ask the vendor: Which AI model and provider power the AI Receptionist, and will you notify us in writing before you change it?

A firm can use the real AI Receptionist on a free plan (25 calls a month, no end date) and paid plans are month-to-month. A credit card is required at signup but is not charged until you authorize it. The Quality Studio lets you run test calls before going live. No NDA or annual contract is needed to try it.

• “Currently, each Free Plan account receives twenty-five (25) Free Calls per month, resetting on the monthly anniversary of your sign-up date.” Smith.ai Terms of Service
• “Smith.ai offers "Receptionist Service Plans" and rates on a month-to-month subscription basis” Smith.ai Terms of Service
• “No commitment, and free forever until you need >25 calls/month.” AI Receptionist product page

Not found in public documents

The amended rule allows use of anonymized or aggregated data, but that does not help here. Neither the Terms nor the Privacy Policy says whether call recordings, transcripts, or caller details are used to train Smith.ai's AI or its model provider's AI. Nothing limits the 'improving our Services' use to de-identified data; the Terms make Smith.ai a joint owner of your identifiable Subscriber Information and allow it to be used to improve the service. The product page also says recordings are used for training without explaining what that means. There is no no-training commitment for either Smith.ai or its AI provider.

• “All Subscriber Information is jointly owned by Smith and you.” Smith.ai Terms of Service
• “relaying messages to you through text and email, identifying spam callers and solicitations, and improving our Services.” Smith.ai Terms of Service

Ask the vendor: Will you confirm in writing that our identifiable call recordings, transcripts, and caller data are never used to train your AI or your AI provider's models?

The Terms give a specific deletion timeline: a response within 3 business days and deletion within up to 30 days. They do not say how long recordings are kept if you never ask, whether backups are purged, or whether you get written confirmation of deletion. Deleting may also mean closing your whole account.

• “We will respond to your written request to delete Subscriber Information within 3 business days.” Smith.ai Terms of Service
• “Deleting Subscriber Information may require deletion of your Smith.ai account and take up to 30 days to execute.” Smith.ai Terms of Service

Not found in public documents

The benefit-of-the-doubt rule only applies when the vendor claims SOC 2, and Smith.ai does not claim SOC 2 anywhere we could find. A recheck on 2026-10-06 found Smith.ai blog posts that tell readers to ask other vendors for SOC 2 reports. A webinar page includes a SOC 2 claim, but it comes from a different company (Speed.ai), not Smith.ai. smith.ai/security and smith.ai/trust return 404 errors. A DPA can be requested by email, so it is available before signing. Smith.ai says it is not HIPAA compliant, so personal injury or medical intake gets no BAA.

• “the terms of the Smith.ai Data Processing Agreement, which are hereby incorporated by reference and a copy of which can be obtained by emailing support” Smith.ai Terms of Service
• “Smith.ai is not HIPAA-compliant, so we cannot handle calls that involve protected health information (PHI).” Smith.ai law firm answering service page

Ask the vendor: Do you have a SOC 2 Type II report covering the AI Receptionist, and can we see it before signing up?

Not found in public documents

Smith.ai says it operates from the United States, and the Privacy Policy names Amazon as the website host. There is no published subprocessor list (including the AI provider or telephony carriers) and no promise to notify customers before adding a new subprocessor.

• “The Platform and Services are controlled and operated from the United States and subject to its laws.” Smith.ai Terms of Service
• “The Website is hosted by Amazon Cloud Services” Smith.ai Privacy Policy

Ask the vendor: Can you send your current subprocessor list, including your AI model provider, and confirm you will notify us before adding new ones?

Addressed, but falls short of the standard

The Terms say each customer's information is kept in a segregated account. Under the amended rule, a vendor SOC 2 claim would count as independent support, but Smith.ai makes no SOC 2 claim. It also has no published penetration test or red-team exercise. The separation promise therefore has no independent support.

• “Smith stores your Subscriber Information in a segregated account that only contains your Subscriber Information.” Smith.ai Terms of Service

Ask the vendor: What independent audit or penetration test confirms that one customer's call data cannot reach another customer, and can we see it?

The Terms commit Smith.ai to report a security breach to you within 48 hours, with all relevant details, and to cooperate on investigation and mitigation. No named breach contact is listed; general contact is support@smith.ai.

• “Smith.ai shall make such report within 48 hours after learning of the Security Breach.” Smith.ai Terms of Service

Addressed, but falls short of the standard

The Terms commit to deleting your information on written request. The only published export guide is filed under the Virtual Receptionists (human) service, not the AI Receptionist, and lists summary fields only. Nothing published confirms that AI Receptionist customers can export their call records, recordings, or transcripts in a usable format. There is no certificate of destruction.

• “Time created, Client name, Client phone, Client address, Client email, Client status, Call disposition, Priority, Summary” Help Center: How to Export Your Smith.ai Call Data via CSV (Virtual Receptionists category)
• “we will delete the requested information and rescind all ownership rights to the Subscriber Information that was deleted.” Smith.ai Terms of Service

Ask the vendor: Can AI Receptionist customers export all call records, full transcripts, and recordings in a standard format before we leave, and will you confirm deletion in writing?

Staff access is limited to people working on your account, and staff sign NDAs. Nothing says staff need your permission each time, that access is logged, or that you can revoke it.

• “This allows us to only allow Smith staff working on your account to access your Subscriber Information.” Smith.ai Terms of Service
• “must sign a Non-Disclosure Agreement requiring them to keep all Subscriber Information confidential” Smith.ai Terms of Service

Ask the vendor: Is staff access to our call data logged, and can we approve or revoke it?

Every AI call is recorded and transcribed and viewable in the dashboard. No published source confirms that AI Receptionist transcripts can be exported by the customer; the only export guide covers the human receptionist service and lists summary fields.

• “Every call is recorded and transcribed, providing a verifiable record for quality assurance, training, and business documentation.” AI Receptionist product page

Ask the vendor: Can we export full transcripts and recordings of every AI call, not just summaries?

The AI talks to callers, qualifies leads, books appointments, and sends texts on its own during live calls, with no person reviewing each call first. Smith.ai describes testing before go-live, customer approval of suggested script changes, and rules for handing calls to human agents. There is no published description of how a firm pauses or turns off the AI immediately.

• “test exactly how it will handle your calls before you turn it on.” AI Receptionist product page
• “you get a recommendation for a simple fix that you can easily approve” AI Receptionist product page

Ask the vendor: How do we immediately pause the AI Receptionist, and what can it do (book, text, transfer) without our approval?

Smokeball's help center names the models behind Archie: OpenAI GPT models (GPT-5.2 is the default) and Anthropic Claude models. A Smokeball blog post also mentions Google Gemini. The Terms say service announcements and changes to the Terms go to the firm's main contact email, and that Smokeball will not make updates that take away a substantial or material benefit during the current term. The Terms do not say how much notice comes before a change, and Smokeball can change the default model.

• “OpenAI GPT-5.2 (default): Balanced, fast, and reliable for everyday work” Configure Archie Settings
• “A dedicated email address (Main Firm Contact) will be utilized for important service announcements... and any changes to our Terms.” Smokeball Terms of Service, clause 35

Not found in public documents

The Terms say Smokeball may offer a free trial, but no official page offers one or gives a length. The pricing page asks visitors to request a quote. Subscriptions run for a set term and auto-renew as 12-month subscriptions. Month-to-month 'Flex Users' are described as extra users at higher rates, not as a way to try the product.

• “Smokeball may offer you a free trial period during which you may use the Services (or a part of them) without paying any Subscription Fees.” Smokeball Terms of Service, clause 10

Ask the vendor: Can our firm get hands-on access to Archie on real or sample matters before signing an NDA or an annual contract, and for how long?

Smokeball's help center says client data is not used to train Smokeball's models or any third party's models, and that Archie runs under zero-data-retention agreements with the AI providers. The security page repeats that customer data is not given to third parties for AI training. Both halves are covered. The Terms separately allow Smokeball to use 'Usage Data', which they define as not being client data; under the amended rule this is acceptable and is noted only for awareness.

• “Client data is never used to train AI models - neither Smokeball's nor any third party's.” How Smokeball AI keeps your data private and secure
• “Archie operates under enterprise agreements with a strict zero-data-retention policy, so data is not stored, retained, or used for training.” How Smokeball AI keeps your data private and secure
• “Smokeball does not provide customer data to third parties for AI model training or third-party AI modelling.” Smokeball Security Policy

Addressed, but falls short of the standard

The Terms say Smokeball 'may' delete account data 30 days after termination, which is permission to delete, not a promise to delete. The Privacy Policy gives no retention periods. Nothing published covers backup deletion schedules, cryptographic erasure, or a certificate of destruction.

• “we may immediately deactivate Your Services and may delete Your account and data after thirty (30) days.” Smokeball Terms of Service, clause 14
• “The specific retention periods depend on the nature of the personal data and why it is collected and processed.” Smokeball Privacy Policy

Ask the vendor: Will you commit in writing to delete all our data, including backups, within a set number of days after we leave, and give us written confirmation when it is done?

Not found in public documents

The benefit-of-the-doubt rule does not help here, because Smokeball never claims a SOC 2 report of its own. Its security page mentions SOC 2 only as something Amazon Web Services (its host) holds, and as a framework its access controls are 'aligned with'. A recheck of smokeball.com on 2026-10-06 found no Smokeball SOC 2 claim. Smokeball relies on ISO 27001:2022, which does not substitute for SOC 2 under this scorecard. No Data Processing Agreement is published or listed as requestable, and no pre-signing BAA was found, which matters for firms holding medical records.

• “Smokeball is proud to be an ISO 27001 certified organization” Smokeball Security Policy
• “Smokeball is certified to ISO 27001:2022 and is independently penetration-tested every year” How Smokeball AI keeps your data private and secure

Ask the vendor: Do you have your own SOC 2 Type II report covering Smokeball and Archie, and can we review it, your Data Processing Agreement, and a BAA (if we hold medical records) before signing?

Addressed, but falls short of the standard

Smokeball says it is hosted on Amazon Web Services and names the countries where its servers operate, but does not say which country holds a US firm's data. The Privacy Policy points to 'product documentation' for sub-processors, but no subprocessor list was found on the website or in the help center. Nothing commits to notifying customers before a new subprocessor is added.

• “Our servers and the Smokeball Group companies are located in the United Kingdom, the United States, and Australia.” Smokeball Privacy Policy
• “applicable sub-processors will be identified in our product documentation.” Smokeball Privacy Policy

Ask the vendor: Can you send your full subprocessor list (including the AI model providers), confirm where our data is stored, and tell us how much notice you give before adding a new subprocessor?

Addressed, but falls short of the standard

Smokeball clearly states that each firm's data is kept logically separate. Smokeball makes no SOC 2 claim of its own, so the Q5 benefit of the doubt does not apply. The yearly penetration test is described only in Smokeball's own words, and no report or tester's summary is published. The amended rule requires a published pen test, so this stays a Fail. It is a close call; a tester's letter would likely turn it into a Pass.

• “Each firm's data is logically isolated. Archie AI can only access your firm's data and cannot access or learn from other Smokeball firms.” Smokeball Security Policy
• “independently penetration-tested every year - most recently in October 2025, with no critical, high, or medium issues found” How Smokeball AI keeps your data private and secure

Ask the vendor: Can you share the testing firm's summary of your most recent penetration test, or a SOC 2 Type II report, confirming that one firm's data cannot be reached from another firm's account?

Not found in public documents

Neither the Terms of Service, the Privacy Policy, the security page, nor the AI help article commits to a time frame for notifying customers of a data breach. The security page lists infosec@smokeball.com as a general security contact, but no breach notice commitment is published.

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering a breach affecting our data, and who will contact us?

Addressed, but falls short of the standard

On request, Smokeball provides a 'Data Backup', but it is limited to documents and basic matter and contact data, and no file format is stated. After termination, the firm has 30 days to retrieve its data. The Terms say Smokeball 'may' delete data after 30 days but do not promise destruction or proof of it.

• “Data Backup includes documents and basic matter and contact card data.” Smokeball Terms of Service, clause 1
• “Smokeball shall grant You access to the Smokeball Subscription for thirty (30) days after the expiration or termination” Smokeball Terms of Service, clause 15

Ask the vendor: What exactly is included in your exit data export and in what formats, and will you confirm in writing that all our data has been destroyed afterward?

The Terms let Smokeball staff view client data for support, document automation, training and other reasons, and the list is open-ended. Nothing published requires the firm's permission first, describes logging of staff access, or lets the firm revoke it.

• “Smokeball may need to access and view Your Client Data including, but not limited to, the following circumstances” Smokeball Terms of Service, clause 18

Ask the vendor: Do your staff need our explicit permission before viewing our client data, is that access logged, and can we revoke it?

Smokeball says authorized users can see a log of questions asked of Archie. It does not say whether the log includes Archie's answers, ties each entry to a user, or can be exported.

• “A log of Archie questions is available to authorized users, supporting visibility and best-practice prompting across your firm.” How Smokeball AI keeps your data private and secure

Ask the vendor: Does the Archie log capture every prompt and answer by user, and can we export it?

For drafting in Word, Smokeball says Archie applies suggested edits as tracked changes that the lawyer can accept, reject, or modify. A third-party news report (LawNext, July 2026) says the newer agentic Archie creates tasks, memos and calendar events with user approval. No official Smokeball document found lists what Archie can do on its own or describes a firm-wide stop control.

• “Rather than automatically rewriting your document, Archie presents suggested edits for your review.” Archie AI for Word

Ask the vendor: Which Archie actions can happen without a user approving them first, and how can our firm turn those features off?

Spellbook names OpenAI and Anthropic as its model providers on its official subprocessor list (Google Cloud is also listed for LLM services). The Terms of Service commit to prompt notice of revisions that materially affect features, on a commercially reasonable efforts basis.

• “OpenAI • Large Language Model (LLM) Provider” Trust Center Subprocessors
• “Anthropic • Large Language Model (LLM) Provider” Trust Center Subprocessors
• “Spellbook will use commercially reasonable efforts to provide Customer with prompt notice of any revision that materially affects features or functionality” Terms of Service s.2.5

A self-serve 7-day free website trial is available, and the Help Center says website trials are not charged automatically. Sales-assisted trials may convert to paid automatically unless cancelled through the sales rep, so use the website trial.

• “Lawyers and legal teams can try Spellbook free for 7 days to test features, workflows” Pricing page
• “Your trial will automatically expire after 7 days” Help Center: Cancelling your trial

Both halves are covered under the amended rule. The Terms bar OpenAI and Anthropic from training on your data, backed by zero data retention agreements. On Spellbook's own side, the Terms limit use of customer data to providing the service. Product improvement uses de-identified and aggregated data, and the calibration clause applies de-identification first and stays specific to your firm. Be aware that Spellbook may keep identifiable inputs and outputs for up to 90 days for quality review, including evaluating system improvements. Nothing published says this data is used for training.

• “will not be used to train, improve, or develop the AI models of those Third Party LLMs” Terms of Service s.4.1(b)(iv)
• “Spellbook may also use de-identified and aggregated Usage Data for general product improvement and analytics” Terms of Service s.4.1(c)(ii)
• “We ensure data privacy with Zero Data Retention agreements, preventing data use for training.” Pricing page

Vendor's own documents contradict each other

There is one specific timeline: quality-review copies of inputs and outputs are deleted within 90 days. For the firm's main stored data, the Trust Center says it is purged when customers leave, by unspecified best practices, while the Help Center says old projects come back when you resubscribe. No deletion timeline, backup schedule, or certificate of deletion is published for stored customer data.

• “retained for no longer than 90 days from the date of collection, then permanently deleted” Terms of Service s.4.1(c)(iv)
• “purges or removes customer data containing confidential information from the application environment, in accordance with best practices, when customers leave the service” Trust Center Controls
• “your previous projects and account data reappear once you resubscribe or upgrade from trial to paid” Help Center FAQ

Ask the vendor: How many days after we cancel are our stored projects, documents, and backups deleted, and will you give us written confirmation when that is done?

The SOC 2 half passes. Spellbook lists a SOC 2 Type II report in its Trust Center, and we assume it is current and covers the product. The DPA half fails. No customer DPA is published on the website, and none is listed as requestable in the Trust Center. The only DPAs listed are Spellbook's own agreements with OpenAI and Anthropic. The Terms of Service include data terms but are not a standalone DPA. A BAA is not required for a contracts tool, but the Trust Center FAQ says Spellbook can agree to one. Rule update: a DPA is no longer required to pass. If a vendor has one, it should make it available on request, so a firm should ask for it before signing.

• “Auditor engagement letter for SOC 2 Type II audit March 1, 2025 - March 1, 2026” Trust Center Resources
• “Spellbook has executed a data processing agreement with the applicable Third Party LLMs” Terms of Service s.4.1(b)

Addressed, but falls short of the standard

The subprocessor list is published with locations (38 vendors, Canada and US), and data location is stated. You can subscribe to Trust Center updates, but no document commits to telling customers before a new subprocessor is added or gives an objection window. The Terms say only that the list is updated from time to time.

• “Primary hosting provider for the Spellbook application, file data and other services.” Trust Center Subprocessors
• “Spellbook uses cloud providers with data centers in Canada and US for storing and processing customer data.” Trust Center FAQ
• “identified in Spellbook's trust center, as updated from time to time” Terms of Service s.4.1(a)

Ask the vendor: Will you commit in writing to notify us a set number of days before adding any new subprocessor that handles our data, with a right to object?

The Terms commit that your data will not be visible or accessible to other Spellbook customers. A claimed SOC 2 Type II and a June 2026 third-party penetration test provide the independent support. Both are available on request through the Trust Center.

• “will not be visible or accessible to other Spellbook customers” Terms of Service s.4.1(c)(i)
• “Auditor engagement letter for SOC 2 Type II audit March 1, 2025 - March 1, 2026” Trust Center Resources
• “Third party penetration testing report performed in June 2026.” Trust Center Resources

Addressed, but falls short of the standard

The Privacy Policy promises breach notice without undue delay but gives no hour or day limit. The Terms of Service have no breach notification timeline. The published contact channel is security@spellbook.com.

• “we notify you without undue delay after we become aware of such breach, as required by applicable law” Privacy Policy

Ask the vendor: Will you commit in writing to notify us within 72 hours of discovering any breach affecting our data, and name the contact who will reach us?

Vendor's own documents contradict each other

Individual documents can be downloaded from Associate, and the Word Add-In works on files in your own systems. Deletion afterward is contradicted: the Trust Center says data is deleted when customers leave, but the Help Center says old data comes back if you resubscribe. No proof of destruction is offered, and there is no published bulk export of everything stored in Associate.

• “You can download a copy of your revised document using the 'Download' button” Help Center: Associate overview
• “Customer data deleted upon leaving” Trust Center Controls
• “If I resubscribe later, do I get my old data back? Yes” Help Center FAQ

Ask the vendor: When we leave, can we export all our Associate projects, playbooks, and chat history, and will you then delete them by a set date and confirm it in writing?

Staff access is limited to authorized personnel with a business need. Nothing published says staff need your permission to view your data, that access is logged for you to see, or that you can revoke it.

• “accessible only to authorized Spellbook personnel on an as-needed basis” Terms of Service s.4.1(c)(iv)
• “The company restricts privileged access to databases to authorized users with a business need.” Trust Center Controls

Ask the vendor: Do Spellbook staff need our explicit permission to view our documents or prompts, is that access logged, and can we revoke it?

No official source describes a customer-exportable log of every AI prompt and response tied to a user. Admin help articles cover user roles and team permissions only.

Ask the vendor: Can our admin export a log of every AI prompt and answer, showing which user made it and when?

In the core Word Add-In and Associate, AI edits appear as redlines that a lawyer accepts or rejects. The newer Autonomous Contract Management product (early access) can be configured to send and receive email, and its approval and stop controls are described only in general terms.

• “When clicking through the applied redlines, you'll have the option to Accept or Reject the applied changes.” Help Center: Associate overview
• “You stay in control of the judgment calls, the approvals, always have the final say.” Autonomous Contract Management page
• “Provides email sending and intake capabilities for ACM functionality when configured.” Trust Center Subprocessors

Ask the vendor: For ACM and any automated features, what can the AI do without a person approving it, and how do we turn those features off?

Not found in public documents

No Supio page names the AI model or provider behind the product. The website Terms of Service cover only the website, and the product contract is not published, so there is no published promise to notify customers before material changes.

Ask the vendor: Which AI model providers power Supio, and will you commit in writing to notify us before you change providers or make material changes to the service?

Supio publishes that firms can start with a pilot on a handful of their own cases, that it does not require lock-in contracts, and its Trust Center refers to a trial of about two weeks. Supio does not publish whether the pilot is free or what it costs.

• “Many firms begin with a pilot on a handful of cases.” Supio Pricing
• “We don't require lock-in contracts or platform fees.” Supio Pricing
• “Data is securely retained for the duration of the Trial, typically about two weeks” Supio Trust Center FAQ

Not found in public documents

Supio says it never uses your data to train outside or foundation models, which covers the model provider half. Nothing published says Supio does not train its own models on your identifiable data, and its site says the product learns your firm's style and each attorney's voice. The Privacy Policy permits use of anonymous information, which is acceptable under the amended rule and is noted only for awareness. The fail is the silence on Supio's own models.

• “Supio never uses your data to train outside models.” Supio Pricing
• “client data never touches a public model and never trains foundation models.” Why Supio
• “It learns your firm's style and tone, each attorney's voice and judgment, and applies them automatically.” Why Supio
• “We may use and disclose anonymous information for any purpose and to any third party in our sole discretion.” Supio Privacy Policy

Ask the vendor: Will you confirm in the contract that our identifiable inputs and outputs are never used to train or improve Supio's own models, and explain how the product learns our firm's style without doing so?

The Trust Center gives a fixed schedule: after you leave, your data is held for one year and then deleted. It also says data can be destroyed on request but gives no timeframe for that. The one year holding period is long, and nothing published says whether backups follow the same schedule.

• “your data will be securely held for 1 year to facilitate a possible return” Supio Trust Center FAQ
• “After this period, if you do not return, your data will be securely deleted.” Supio Trust Center FAQ

Supio claims SOC 2 Type II, which now counts as current under the benefit of the doubt. However, Supio does not publish a DPA or list one as requestable in its Trust Center. The product processes medical records in PI files, and Supio does not publish a BAA or confirm one is available before signing. A HIPAA compliance claim is not the same as a BAA offer. Rule update: a BAA is required only when the product handles HIPAA-covered health information. Most plaintiff firms are not HIPAA covered entities, so this is not counted against the vendor. A firm that acts as a business associate of a healthcare provider should still ask for a BAA.

• “Supio is SOC 2 Type II certified and uses Vanta to provide customers with 24/7 continuous monitoring” Supio Security page
• “Supio is HIPAA and GDPR compliant and SOC2 Type2 certified.” Supio Pricing

Not found in public documents

Supio publishes where data is stored and names AWS and Azure as hosts. It does not publish a full subprocessor list, including any AI model provider, and does not say it gives notice before adding subprocessors. The Privacy Policy says data is stored only in the United States, which differs from the other location statements.

• “Our cloud infrastructure is hosted on Amazon Web Services and Azure within the United States” Supio Trust Center FAQ
• “Your data will remain in your country of origin, with data centers in the U.S., UK, and Canada.” Supio Security page

Ask the vendor: Can you send your full subprocessor list, including AI model providers, and confirm in writing how much notice you give before adding one?

Not found in public documents

Supio's SOC 2 Type II claim and annual penetration testing would now count as independent support. The fail is that no published source states customer data is isolated or logically separated between firms. Saying data is never shared outside your firm is a sharing promise, not a separation statement.

• “We never share it outside your firm.” Supio Pricing
• “The company's penetration testing is performed at least annually.” Supio Trust Center Controls

Ask the vendor: How is our data kept logically separated from other firms' data, and will you state that in writing?

Not found in public documents

None of the reviewed documents gives a breach notification deadline. The only security contact listed is security@supio.com.

Ask the vendor: Will you commit in writing to notify us of a data breach affecting our data within 72 hours, and who will contact us?

Not found in public documents

Supio describes destroying data, but no published source describes how a firm exports its data or in what format.

• “Data is securely destroyed either upon reaching the end of its retention period or upon your request” Supio Trust Center FAQ

Ask the vendor: How do we export all our case data and work product, in what formats, and will you give written confirmation once it is destroyed?

Supio says privileged access to its databases and production systems is limited to staff with a business need, and that employees are background checked. It does not say staff need your explicit permission, that their access is logged, or that you can revoke it.

• “The company restricts privileged access to databases to authorized users with a business need.” Supio Trust Center Controls
• “Every Supio employee goes through extensive background checks” Supio Security page

Ask the vendor: Do Supio staff need our explicit permission to access our data, and is that access logged and revocable by us?

Nothing published describes a log of AI prompts and answers, or a way to export one.

Ask the vendor: Can we export a log of every AI prompt and answer, tied to the user who ran it?

Supio markets an agentic platform that produces demand letters and litigation drafts and says outputs link back to their sources. Nothing published describes approval steps before output goes out, limits on what the agent can do by itself, or a way to stop it.

• “Every output traces back to its source, so attorneys can stand behind any fact in court.” Why Supio

Ask the vendor: What can the Supio agent do without a person approving it, and how can we require approval or stop it?

Not found in public documents

No official Tavrn page names the underlying AI model or provider. The website and security page describe only Tavrn's own AI and AWS hosting. The Main Services Agreement promises reasonable efforts at advance notice of scheduled service disruptions, but nothing about changes to the AI model.

Ask the vendor: Which AI model provider powers Tavrn's chronologies and demand letters, and will you notify us in writing before switching providers?

Addressed, but falls short of the standard

Product pages show a 'Start Free Trial' button, but every such button links to the demo request page, which is a form to book a call with the sales team. No published page describes what the trial includes, how long it lasts, whether it gives hands-on access to the real product, or whether it requires an NDA or contract commitment.

• “Start Free Trial” Tavrn Demand Letters product page
• “Schedule a time with our team” Tavrn Request Demo page

Ask the vendor: Can our firm get hands-on access to the real Tavrn product on our own cases before signing an NDA or an annual contract, and for how long?

The homepage says customer data is never used to train, fine-tune, or improve any AI models, which covers both Tavrn's own models and its AI provider's. Section 3.3 of the contract lets Tavrn analyze information about how the service is used and performs, including information concerning customer data, to improve its services, and to share it only in aggregate or de-identified form. Under the amended rule this is a usage-data clause, not a license to train on your case files. Worth knowing: the clause does not say Tavrn's internal analysis is limited to de-identified data, and the AI provider is not named.

• “Your data is never used to train, fine-tune, or improve any AI models” Tavrn homepage
• “data and other information relating to the provision, use and performance of various aspects of the Services” Tavrn Main Services Agreement, Section 3.3
• “disclose such data solely in aggregate or other de-identified form in connection with its business” Tavrn Main Services Agreement, Section 3.3

Addressed, but falls short of the standard

The Trust Center lists 'customer data deleted upon leaving' and data retention procedures as controls. No timeframe, backup schedule, erasure method, or certificate of destruction is published.

• “Customer data deleted upon leaving” Tavrn Trust Center

Ask the vendor: How many days after we ask or after termination will our data, including backups, be deleted, and will you give us written confirmation?

Tavrn claims a SOC 2 Type 2 report in its Trust Center, which now counts as current. The fail is on paperwork. The Trust Center lists six requestable documents (security policies, SOC 2 Type 1 and Type 2 reports, a HIPAA attestation), and none of them is a DPA or a BAA. The website and contract publish neither. Tavrn processes medical records, so a BAA is required, and none is confirmed as available before signing. A HIPAA attestation is not a BAA. Rule update: a BAA is required only when the product handles HIPAA-covered health information. Most plaintiff firms are not HIPAA covered entities, so this is not counted against the vendor. A firm that acts as a business associate of a healthcare provider should still ask for a BAA.

• “SOC 2 Type 2 Report ... Request access” Tavrn Trust Center resources
• “HIPAA Attestation ... Request access” Tavrn Trust Center resources

Not found in public documents

Tavrn names AWS as its cloud host. No subprocessor list for the product, no promise to notify before adding subprocessors, and no data location are published. The privacy policy covers only the website.

• “Tavrn uses AWS as its cloud provider” Tavrn homepage

Ask the vendor: Can you send your full list of subprocessors (including AI model providers), tell us where our data is stored, and explain how you notify customers before adding a new one?

Not found in public documents

Tavrn's SOC 2 Type 2 claim would count as independent support, but there is nothing for it to support. No official source we checked (homepage, security page, Trust Center overview and controls, Main Services Agreement) says one firm's data is isolated or kept logically separate from another's.

Ask the vendor: How do you keep our firm's data separate from other customers' data, and does your SOC 2 report cover that control?

Not found in public documents

No published document gives a deadline for notifying customers of a data breach. The only contact listed on the website is contact@tavrn.ai.

Ask the vendor: Within how many hours will you notify us of a security breach involving our data, and who will contact us?

Not found in public documents

Deletion when a customer leaves appears as a Trust Center control. No official source explains how a firm can export its data, or in what format, when it leaves.

• “Customer data deleted upon leaving” Tavrn Trust Center

Ask the vendor: If we leave, how do we export our chronologies, demands, and records, in what formats, and will you confirm deletion afterward?

The Trust Center lists general access-control items. Nothing public says whether Tavrn staff need the firm's permission to view case data, or whether that access is logged or can be revoked.

Ask the vendor: Can Tavrn staff view our case files without our permission, and is that access logged and revocable by us?

Nothing public describes a log of AI prompts and outputs, or any export of such a log.

Ask the vendor: Can we export a log of every AI request and output, showing which user ran it?

Tavrn markets medical record retrieval and AI-drafted demand letters. Nothing public describes approval steps before records requests or drafts go out, or a way for the firm to stop automated actions. The contract warns only that AI output may contain errors.

• “The Output of Tavrn is AI- generated, and may contain errors and misstatements or may be incomplete.” Tavrn Main Services Agreement, Section 1.3

Ask the vendor: Does an attorney at our firm approve each records request and demand letter before it is sent, and how do we pause or stop automated actions?

Thomson Reuters names OpenAI (GPT) and Google (Gemini) as its AI model providers on its CoCounsel page and in its CoCounsel subprocessor list. The CoCounsel Legal product page itself does not name a model. The US General Terms promise six months' notice before discontinuing material parts of the service, and say routine updates may not materially reduce overall functionality. There is no specific promise to notify you before switching AI models; routine updates can happen without notice. The general change clause meets this test, but only just.

• “We process all interactions with OpenAI GPT and Google Gemini in the U.S.” CoCounsel overview and data security FAQ
• “Open AI ... LLM Provider; Google LLC ... Cloud Infrastructure, LLM Provider” CoCounsel Core 2.0 Subprocessor List
• “Thomson Reuters may discontinue Services in its entirety or material components thereof with at least six (6) months' prior notice” General Terms and Conditions, US Version 6.0, Section 17.2

Not found in public documents

The pricing page offers 1, 2, and 3 year subscriptions only. The CoCounsel Legal page offers a demo request and an interactive demo, but no published free trial, sandbox, or month-to-month plan. Free trial request pages exist for CoCounsel Core and CoCounsel Essentials, which are different products. The US General Terms say agreements automatically renew for one-year terms.

• “Longer-term subscriptions typically offer greater savings.” CoCounsel Legal plans and pricing

Ask the vendor: Can our firm get hands-on access to CoCounsel Legal itself, with our own test matters, before signing an NDA or an annual contract?

The contract covers both halves. Section 3.6 of the US General Terms bars Thomson Reuters and its third-party AI providers from training AI models on your data (including prompts) or outputs unless you agree otherwise in writing. The CoCounsel page adds that OpenAI and Google are contractually barred from training and use zero-retention settings. Something to be aware of: Section 3.5 lets Thomson Reuters use prompts and outputs for quality and safety testing, with automated safeguards meant to exclude sensitive information. That is testing, not training. A training allowance in the Legal Product Specific Terms applies to a different product (Document Intelligence), not CoCounsel.

• “Thomson Reuters will not, and will not permit its third-party providers to, develop, train, or fine-tune any generative artificial intelligence models” General Terms and Conditions, US Version 6.0, Section 3.6
• “Your user content and prompts are not used to train or improve CoCounsel and associated products or LLMs” CoCounsel overview and data security FAQ

Two specific mechanisms are published: deletion of live data within 90 days after termination on written request, and written confirmation of destruction on request. Users can also delete matters in the product. Caveats: backup copies are deleted only in the normal course of business with no stated schedule, and the DPA allows retention required by law or by Thomson Reuters' own data retention policies.

• “Thomson Reuters will delete Customer Data hosted in the production environment within ninety (90) days after termination.” General Terms and Conditions, US Version 6.0, Section 17.6
• “Thomson Reuters will securely destroy and, upon request, confirm the destruction of all copies of Your Data” Data Security Addendum v2.0, Section 2.12

Thomson Reuters' CoCounsel Legal security profile lists SOC 2 Type 2 reports for CoCounsel Core (January 2026), Westlaw and Practical Law, with bridge letters through October 2026. Under the amended rule this is treated as current and covering the product. Full reports are available on request through the profile. The DPA and Data Security Addendum are published on the website, so they are available before signing. No BAA is published. CoCounsel Legal is a general research and drafting tool, not a health-information product, so this does not affect the result, but firms that upload medical records may want to ask.

• “Thomson Reuters - Cocounsel Core - SOC 2 Type 2 Report - January 2026.pdf” CoCounsel Legal security profile (Whistic)
• “SOC 2 reports and Bridge Letters applicable to CoCounsel Core, Practical Law, and Westlaw.” CoCounsel Legal security profile (Whistic)
• “This DPA shall be effective for all contracts and renewals entered into on or after February 1, 2025.” Thomson Reuters Data Processing Addendum Version 3

Addressed, but falls short of the standard

A subprocessor list is published for CoCounsel Core 2.0 (February 2025), but there is no separate list for CoCounsel Legal as a whole. The list gives company headquarters, not where data is processed. Data location is stated on the CoCounsel page (US hosting, with UK, Australia, and Canada options on request). The DPA says updates are posted on webpages and gives a right to object, but it does not promise to notify customers before a new subprocessor is added.

• “Amazon Web Services, Cloudflare, Data Dog, Fivetran, Google LLC, Microsoft Corporation, Open AI, Pendo.io Inc, Snowflake, Syncly” CoCounsel Core 2.0 Subprocessor List
• “Thomson Reuters' Subprocessors, and updates thereof, can be found on its webpages here or in other notices provided from time to time.” Data Processing Addendum v3, Section 3.7.1
• “Systematic procedures and hosting are in the U.S.” CoCounsel overview and data security FAQ

Ask the vendor: Will you notify us in advance, and how many days ahead, before adding or replacing any subprocessor for CoCounsel Legal, and can you give us the complete current list for CoCounsel Legal?

The Data Security Addendum commits to keeping each customer's data logically separated from other customers' data. The SOC 2 Type 2 report listed on the security profile counts as current independent support under the amended rule. The Data Security Addendum also describes periodic penetration testing.

• “Your Data hosted within the Services in a production environment is maintained so as to preserve logical segregation of Your Data from data of others.” Data Security Addendum v2.0, Section 2.11
• “Thomson Reuters - Cocounsel Core - SOC 2 Type 2 Report - January 2026.pdf” CoCounsel Legal security profile (Whistic)

The Data Security Addendum commits to notifying you within 72 hours of discovering a security breach. No breach contact is named. The DPA lists privacy.enquiries@tr.com for privacy questions.

• “without undue delay but in any event within seventy-two (72) hours of discovery, notify Customer of a Security Breach” Data Security Addendum v2.0, Section 4.1

At termination you can get a copy of your data within 30 days, or download it yourself beforehand. Thomson Reuters then deletes live data within 90 days on request, and the Data Security Addendum lets you ask for written confirmation of destruction. The termination clause says Thomson Reuters standard format without naming a file type. In-product downloads of AI work product in PDF, Word, and Markdown are documented.

• “provide Customer with a copy of Customer Data hosted in the production environment in a Thomson Reuters standard format within thirty (30) days” General Terms and Conditions, US Version 6.0, Section 17.6
• “users can choose the exact file format and citation style they want, including two new formats, PDF and Markdown” CoCounsel Legal June 2026 Releases

Thomson Reuters staff access is limited to authorized personnel, and the AI providers' human review is turned off. Nothing published says staff need your permission to view your data, or that you can see or revoke their access.

• “Policies to restrict access to Your Data only to authorized Thomson Reuters personnel and subcontractors” Data Security Addendum v2.0, Section 1
• “systemic controls to turn off third parties' abuse-monitoring solutions to ensure there is no human access” CoCounsel overview and data security FAQ

Ask the vendor: Do Thomson Reuters staff need our explicit permission before viewing our CoCounsel Legal data, is every access logged, and can we review and revoke that access?

Admins get a usage analytics dashboard showing chat threads, messages sent, and skills run, broken down by user, and the table data can be exported. The help page describes activity counts only and does not mention prompt or answer content. Nothing published describes an admin-exportable log of every prompt and answer.

• “insights into each users' activity” CoCounsel Legal help: Usage analytics

Ask the vendor: Can our firm administrator export a complete log of every CoCounsel Legal prompt and response, tied to the user who made it?

CoCounsel Legal uses agentic AI that plans and carries out multi-step research and drafting. Output goes back to the user, and nothing published says it sends messages or files anything. Westlaw Brief Builder is described as pausing for attorney input at key decision points. Other features have no published approval gates, list of what the AI does without approval, or stop control. The terms place review responsibility on the customer.

• “it pauses at key decision points and waits for your input” The next generation of CoCounsel Legal: 2026 updates
• “Output must be reviewed by a licensed or qualified professional before reliance, use, or distribution.” General Terms and Conditions, US Version 6.0, Section 3.3

Ask the vendor: Which CoCounsel Legal agentic features run multiple steps without stopping for our approval, and how can a user or admin stop a running task?

Viktor names its model providers (Anthropic, OpenAI, Google) on its security page and in the Privacy Policy. The General Terms promise 30 days' notice of material changes, and the DPA requires email notice before new subprocessors (including AI providers) are added. Note the Definitions say changes to specific AI models do not count as discontinuing the core service.

• “Inference runs on OpenAI, Anthropic, and Google.” Viktor Security page
• “Provider will provide notice at least thirty (30) days before the changes take effect” General Terms of Service
• “Zeta AI shall notify Customer of any intended addition or replacement of a Sub-Processor by email to Customer's account administrator” Data Processing Agreement, Section 7.4

Anyone can sign up and use the real product with $100 in free credits, no credit card and no sales call. Paid Team plans start at $50 per month and are shown with monthly billing, with no annual commitment listed.

• “$100 in credits on the house. No credit card, no sales call, no catch.” Viktor Security page
• “Start free with $100 in credits” Viktor Pricing

Viktor states it does not train Viktor on customer data, and that its model providers (OpenAI, Anthropic, Google) operate under no-training agreements for Viktor traffic. The Privacy Policy repeats both promises. The General Terms let Viktor use aggregated, de-identified, or anonymized data to improve the service, which is acceptable under the amended rule. Worth knowing: the contract wording is narrower than the marketing (it names 'foundation' or 'general-purpose' models).

• “We do not use customer data to train Viktor, and our model providers do not either.” Viktor Security page
• “all three operate under a no-training, zero-retention agreement for Viktor traffic.” Viktor Security page
• “We do not train our own or third-party foundation models on Customer Data, and we do not permit AI providers to do so.” Privacy Policy
• “Provider may collect, generate, and use aggregated, de-identified, or anonymized data derived from Customer's use of the Service” General Terms of Service, Section 2.3

The Privacy Policy says production data is deleted about 30 days after a request and backups roll off on about a 35-day cycle. The DPA promises written confirmation of deletion if you ask within 60 days of deletion. The security page gives a different timeline (7 business days for full account deletion). See contradictions.

• “Remaining copies are removed as encrypted backups age out on their normal rotation (currently ~35 days)” Privacy Policy, Section 6
• “Zeta AI shall provide written confirmation that Customer Personal Data has been deleted from Zeta AI's systems” Data Processing Agreement, Section 9.2

Addressed, but falls short of the standard

Viktor's own security page says it holds SOC 2 Type I and that Type II is still in progress. Even with the benefit of the doubt on dates, a Type I claim with Type II in progress still fails under the amended rule. A DPA is published and applies automatically. No BAA is offered, and the Acceptable Use Policy bars regulated data such as health information.

• “SOC 2 Type 1 | Certified” Viktor Security page
• “Independent attestation that our security controls operate as designed. Type II in progress.” Viktor Security page
• “As of 9 October 2025, the Viktor system has been subject to an independent SOC 2 examination” Data Processing Agreement, Section 4.5

Ask the vendor: When will your SOC 2 Type II audit be completed, and what audit period will it cover?

The subprocessor list (including Anthropic, OpenAI, Google and other AI providers) is published in the Privacy Policy. The DPA requires email notice to the account administrator before adding a subprocessor, with a 30-day objection window. Data is hosted in the US. EU hosting is Enterprise only. AI providers may temporarily retain data under their own API retention policies.

• “core language models (accessed directly and through Google Cloud Vertex AI and Amazon Bedrock)” Privacy Policy, Section 4 (subprocessor table)
• “thirty (30) days from the date of such notice, provided that the objection is based on reasonable data protection grounds” Data Processing Agreement, Section 7.4
• “Customer Data is stored with reputable cloud service providers in U.S. regions” Privacy Policy, Data center location

Addressed, but falls short of the standard

Viktor clearly says each customer workspace is isolated. The only independent support is a SOC 2 Type I report, which does not count under the amended rule. No published pen test or described red-team exercise was found.

• “Every workspace runs in a sandboxed execution environment with no cross-tenant access.” Viktor Security page
• “SOC 2 Type 1 | Certified” Viktor Security page

Ask the vendor: Can you share a current independent penetration test or SOC 2 Type II report that tests isolation between customer workspaces?

The DPA commits to notifying the customer of a breach within 72 hours. The Product Terms point to the DPA for timing. Security contact: security@viktor.com.

• “without undue delay, and in any event within seventy-two (72) hours of becoming aware of the Security Breach” Data Processing Agreement, Section 6.1

Addressed, but falls short of the standard

You can ask for your data back before deletion, and deletion with written confirmation is promised. But no document says what format the export comes in (CSV, PDF, native files). The DPA also lets Viktor charge extra fees for data export.

• “Zeta AI shall, at Customer's choice, return or delete Personal Data Processed on behalf of Customer” Data Processing Agreement, Section 9.2
• “Where legally permitted, customers may request an export prior to deletion.” Privacy Policy, Exports
• “Zeta AI may charge additional fees for any assistance, cooperation, audit support, data export” Data Processing Agreement, Section 3.3

Ask the vendor: In what file format will you return our data on exit, and will there be a fee for the export?