Harvey
harvey.ai
Automated, point-in-time. Not independently reviewed or approved by TRUSTYCYBER unless expressly stated.
- No clear commitment that your data will not train their models
See Before you sign, with what to ask for ↓
Scanned Oct 21, 2026 · Public evidence · Point-in-time
Higher scores indicate stronger, independently supported AI assurance, governance, AI systems, models, data, supply chain and third-party assurance evidence. The score and the decision are related but different: a well-scored vendor can still carry conditions.
Before you sign
Why it matters: The public sources scanned do not clearly state that customer prompts, files, and outputs are excluded from training or fine-tuning of first- or third-party models.
What to ask for: Get a no-training clause covering first- and third-party models into the DPA/order form.
“We use publicly available information (for example, judgments, decisions, public filings, etc.,) to develop, train, and improve our AI platform”
“Harvey contractually prohibits model providers from training on customer data and only uses your data for processing your requests, not for model improvement. Harvey requires Zero Data Retention (ZDR) by model providers.”
“Harvey contractually guarantees through our Platform Agreement that your data stays yours. We don’t use inputs, outputs, or uploaded documents to train underlying models.”
“Yes, but only if you explicitly request it. In that case, a bespoke model is created exclusively for you, and your data is never used to train models used by other customers.”
“When Harvey processes Personal Data contained in Account Information and Usage Personal Data, Harvey is acting as a Data Controller, and Harvey will: 15.1 Process such Personal Data, in compliance with applicable Data Protection Law, and strictly to: 15.1.1 manage Your account, and Harvey’s relationship with You, including identity verification, and providing support, required to access or use the Service; 15.1.2 to carry out Harvey’s core business operations, such as accounting, and auditing; 15.1.3 to prevent, detect, or investigate security incidents and manage the security of Harvey's platform and Service; 15.1.4 for business analytics and product strategy; and 15.1.5 to develop and improve new products and services and improve the performance, functionality, safety, and security of the services.”
Buyer questionsanswered only from the public evidence scanned, click a row for the answer
!Will they train on your data?Ask the vendor
Harvey states it uses publicly available information such as judgments, decisions and public filings to develop, train and improve its AI platform; this is a statement about public training sources, not about customer data.
Requires written confirmation, see Before you sign ↓
“We use publicly available information (for example, judgments, decisions, public filings, etc.,) to develop, train, and improve our AI platform”
“Harvey contractually prohibits model providers from training on customer data and only uses your data for processing your requests, not for model improvement. Harvey requires Zero Data Retention (ZDR) by model providers.”
“Harvey contractually guarantees through our Platform Agreement that your data stays yours. We don’t use inputs, outputs, or uploaded documents to train underlying models.”
“Yes, but only if you explicitly request it. In that case, a bespoke model is created exclusively for you, and your data is never used to train models used by other customers.”
“When Harvey processes Personal Data contained in Account Information and Usage Personal Data, Harvey is acting as a Data Controller, and Harvey will: 15.1 Process such Personal Data, in compliance with applicable Data Protection Law, and strictly to: 15.1.1 manage Your account, and Harvey’s relationship with You, including identity verification, and providing support, required to access or use the Service; 15.1.2 to carry out Harvey’s core business operations, such as accounting, and auditing; 15.1.3 to prevent, detect, or investigate security incidents and manage the security of Harvey's platform and Service; 15.1.4 for business analytics and product strategy; and 15.1.5 to develop and improve new products and services and improve the performance, functionality, safety, and security of the services.”
✓How long do they keep your data?Clear
Harvey says each AI provider must commit to no retention of customer data (ephemeral processing then deletion), no human review of customer data, and no AI training on customer data.
“No retention of any customer data: All data is processed ephemerally and then deleted. No human eyes on customer data: We call this “eyes off,” and it means that no human has access to or reviews our customer data. No AI training on customer data: Our customers’ data belongs to them and may not be used to train AI models.”
“Customers define their own data retention periods, with options as short as three hours, allowing them to tailor data handling to their specific risk profiles and internal governance requirements. Beyond retention settings, users can delete their data within the app at any time.”
“Within 30 days of either (i) termination of the Service; or (ii) upon Your reasonable request; Harvey shall, and shall direct each Subprocessor to, return to You or delete the Customer Personal Data, unless Harvey is required by law to retain such Customer Personal Data.”
✓Who else can access your data?Clear
Harvey's blog announced Fable 5.1, an Anthropic model, as live in Harvey from 1 September 2026, consistent with Anthropic appearing on the subprocessor list as an AI provider.
“Fable 5.1, Now Live in Harvey 9/1”
“Microsoft Corp AI provider and cloud services provider US, EU, Switzerland, Australia OpenAI, LLC AI provider US, EU, Switzerland, Australia Google Cloud Platform AI provider US, EU, Switzerland, Australia Amazon Web Services AI provider and cloud services provider US, EU, Switzerland, Australia Anthropic, PBC AI provider US* ElevenLabs Inc. AI provider (including voice services) US, EU DeepL SE AI provider (machine translation) US, EU Mistral AI SAS* AI provider EU Baseten Labs, Inc. AI provider US, EU, Australia Fireworks.ai, Inc. AI provider US, EU, Australia”
“SuSea, Inc (you.com) Web browsing US, EU Parallel Web Systems Inc. Knowledge Sources and Web browsing US, EU RELX Inc., d.b.a. LexisNexis Ask LexisNexis US”
“enter into contractual arrangements with each Subprocessor binding them to data protection obligations that are materially equivalent to, and no less protective than, those imposed on Harvey under this DPA, and to provide the same level of security provided for in this DPA. Harvey will remain fully liable to You for the performance of each Subprocessor”
!Where is your data processed?Ask the vendor
Anthropic is listed for US processing only and Mistral for EU only; customers who chose other Harvey processing locations may still reach those providers' models through Harvey's model hosting platforms such as AWS Bedrock or Google Vertex, so the serving party can differ from the model developer.
Confirm in writing: Ask the vendor to state this in writing before signing.
“*Customers that have selected Harvey processing locations outside the listed locations for Anthropic and Mistral may still access models from such AI providers as available through Harvey’s model hosting platforms (e.g., AWS Bedrock, Google Vertex).”
“Harvey hosts its cloud environment in Microsoft Azure. For customers with requirements or preferences for data localization we offer processing in the EU and Switzerland or Australia. This applies to Harvey subprocessors as well.”
“You acknowledge that, unless You and Harvey have agreed, in Your currently operative order form or otherwise in writing, to Process and store Customer Personal Data exclusively in a different geographic location, You may transfer Personal Data to Harvey in the United States, in order for Harvey to provide the Service and for the purposes set out in this DPA.”
✓What happens in a security incident?Clear
The DPA commits Harvey to notify customers of a Personal Data Breach within the timeframe set in the Security Addendum and to supply the information Data Protection Law requires; the specific timeframe is in the Security Addendum, which was not among the collected documents.
“Harvey will notify You in accordance with the timeframe set out under the heading “Incident Detection and Response” in the Security Addendum which is incorporated into this DPA. Harvey will provide You with all information about the Personal Data Breach as required by Data Protection Law”
Email to send the vendor6 items to confirm in writing
Hello Harvey team, We are assessing Harvey as part of our supplier review. Before we proceed, please confirm the following in writing: 1. Do you use our data (inputs, outputs or uploaded files) to train or improve your models? Please confirm in writing, including any opt-out and whether it applies to every plan. 2. In which countries and regions is our data processed and stored, and can processing be limited to a region we choose? 3. What evaluation suite and acceptance thresholds are applied before a new or updated model is released to customers, and how are customers notified when the model serving a given feature changes? 4. Intercom appears to be involved as a application builder: confirm whether this dependency exists, and whether it processes customer data. 5. Atlassian appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data. 6. Microsoft appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data. A written reply to each point, or a link to where it is documented, is enough. Where a point is covered by a certification or independent report, please include the certificate number or the report and its date. Thank you,
Key findingsclick a row for the evidence
✓Unusually deep independent assurance stack, including AI-specific certificationsStrong
Harvey claims SOC 2 Type II and ISO 27001 (annual, Schellman), ISO/IEC 42001 and AIUC-1 (Schellman as accredited AIUC-1 body), an IRAP attestation, ISO 27701, TX-RAMP and Data Privacy Framework certification, plus penetration testing by NCC Group and Bishop Fox. Every item is vendor-stated on Harvey's own pages; certificates sit behind the gated Trust Center and no registry or auditor source was in the collected documents, so all are recorded as vendor_claimed_only or scope_unclear.
ISO 42001 and AIUC-1 cover the AI management system itself, not just information security, which is rare among legal AI vendors. A buyer can request the certificates and Statements of Applicability through the Trust Center and verify issuer, scope and dates directly.
Question for vendor: Please provide the ISO/IEC 42001 and AIUC-1 certificates with scope statements and issue dates, and confirm whether the IRAP assessment covers the Australian processing region specifically.
“We are pleased to announce that Harvey has recently achieved AIUC-1 certification, ISO 42001 certification, and IRAP attestation, reinforcing our commitment to security, safety, reliability, and responsible AI governance.”
“Announcing Successful Completion of 2026 SOC 2 Type II Audit and ISO 27001 Certification We are pleased to announce that Harvey has successfully completed another SOC 2 Type II audit and renewed our ISO 27001 certification.”
“Harvey partners with top-tier security firms, including Schellman, NCC Group and Bishop Fox, to perform in-depth audits. This offers external validation that Harvey meets the highest standards of resilience.”
“Compliance AIUC-1 CCPA EU AI Act GDPR IRAP ISO/IEC 27001 ISO/IEC 27001 SoA ISO/IEC 27701 ISO/IEC 27701 SoA ISO/IEC 42001:2023 ISO/IEC 42001 SoA SOC 2 Type 2 TX-RAMP VPAT WCAG”
“Harvey has certified to the U.S. Department of Commerce that it adheres to the EU-U.S. Data Privacy Framework Principles (“ EU-U.S. DPF Principles” ) with regards to the processing of personal data received from the European Union (“ EU ”) and UK in reliance on the EU-U.S. DPF and the UK Extension.”
✓Clear no-training commitment with retention controls and provider flow-downStrong
Harvey states, in a contractual guarantee under the Platform Agreement, that inputs, outputs and uploaded documents are not used to train models; model providers are contractually bound to zero data retention, no training and no human review; customers set retention down to three hours and can delete in-app; the DPA requires deletion within 30 days of termination. Training on a customer's own data is opt-in only and yields a bespoke model for that customer.
These are the answers most legal buyers need on training, retention and recipients, and they are stated as contract terms rather than marketing. The security page's 'by default, never trains on customer data' wording and the opt-in bespoke model path mean the buyer should confirm their own tenant has not enabled any training option.
Question for vendor: Confirm in writing that no training or fine-tuning option is enabled for our workspace, and that Zero Data Retention terms are in place with every AI provider on the current subprocessor list, including those reached via AWS Bedrock or Google Vertex.
“Harvey contractually prohibits model providers from training on customer data and only uses your data for processing your requests, not for model improvement. Harvey requires Zero Data Retention (ZDR) by model providers.”
“No retention of any customer data: All data is processed ephemerally and then deleted. No human eyes on customer data: We call this “eyes off,” and it means that no human has access to or reviews our customer data. No AI training on customer data: Our customers’ data belongs to them and may not be used to train AI models.”
“Harvey contractually guarantees through our Platform Agreement that your data stays yours. We don’t use inputs, outputs, or uploaded documents to train underlying models.”
“Yes, but only if you explicitly request it. In that case, a bespoke model is created exclusively for you, and your data is never used to train models used by other customers.”
“Customers define their own data retention periods, with options as short as three hours, allowing them to tailor data handling to their specific risk profiles and internal governance requirements. Beyond retention settings, users can delete their data within the app at any time.”
“Within 30 days of either (i) termination of the Service; or (ii) upon Your reasonable request; Harvey shall, and shall direct each Subprocessor to, return to You or delete the Customer Personal Data, unless Harvey is required by law to retain such Customer Personal Data.”
!Ten AI providers with region-dependent routing; model-to-feature mapping not publishedGap
The subprocessor list names ten AI providers (Microsoft, OpenAI, Google, AWS, Anthropic, ElevenLabs, DeepL, Mistral, Baseten, Fireworks.ai) and Harvey has released its own open-weight model (Tenet). Anthropic is listed for US processing only and Mistral for EU only, with a footnote that customers in other regions may still reach those models through AWS Bedrock or Google Vertex. The collected documents do not say which model serves which Harvey feature or which hosting platform serves it in each region.
For an in-region customer the party actually running inference may be AWS or Google rather than the model developer, which changes the subprocessor whose terms govern the request. A buyer needs the model-to-feature and model-to-hosting mapping to assess the chain properly.
Question for vendor: For our selected processing region, which models serve Assistant, Vault, Knowledge and Agents, and for each, is inference performed by the model developer or by a hosting platform (Azure, Bedroom, Vertex, Baseten, Fireworks)?
“Microsoft Corp AI provider and cloud services provider US, EU, Switzerland, Australia OpenAI, LLC AI provider US, EU, Switzerland, Australia Google Cloud Platform AI provider US, EU, Switzerland, Australia Amazon Web Services AI provider and cloud services provider US, EU, Switzerland, Australia Anthropic, PBC AI provider US* ElevenLabs Inc. AI provider (including voice services) US, EU DeepL SE AI provider (machine translation) US, EU Mistral AI SAS* AI provider EU Baseten Labs, Inc. AI provider US, EU, Australia Fireworks.ai, Inc. AI provider US, EU, Australia”
“*Customers that have selected Harvey processing locations outside the listed locations for Anthropic and Mistral may still access models from such AI providers as available through Harvey’s model hosting platforms (e.g., AWS Bedrock, Google Vertex).”
“Fable 5.1, Now Live in Harvey 9/1”
“Introducing Harvey Tenet, our first post-trained open-weight model.”
!US is the contractual default location; usage data used for product improvementGap
The security page advertises EU/Switzerland and Australia processing, but DPA section 12.1 makes the United States the default unless the order form or other writing fixes a different exclusive location. Separately, DPA section 15 makes Harvey an independent controller of Account Information and Usage Personal Data and permits use of it for analytics and to develop and improve products; that category is defined to exclude Customer Data and Content, but Usage Data as described in the Privacy Policy includes query volume and type.
In-region processing is only binding if written into the order form. The usage-data clause is not a training disclosure, but the buyer should confirm the boundary between usage metadata and prompt or output content so the no-training commitment is not undercut by a definitional gap.
Question for vendor: Confirm that our order form specifies exclusive in-region processing, and confirm in writing that Usage Data used under DPA section 15 never includes prompt text, outputs or document content.
“You acknowledge that, unless You and Harvey have agreed, in Your currently operative order form or otherwise in writing, to Process and store Customer Personal Data exclusively in a different geographic location, You may transfer Personal Data to Harvey in the United States, in order for Harvey to provide the Service and for the purposes set out in this DPA.”
“When Harvey processes Personal Data contained in Account Information and Usage Personal Data, Harvey is acting as a Data Controller, and Harvey will: 15.1 Process such Personal Data, in compliance with applicable Data Protection Law, and strictly to: 15.1.1 manage Your account, and Harvey’s relationship with You, including identity verification, and providing support, required to access or use the Service; 15.1.2 to carry out Harvey’s core business operations, such as accounting, and auditing; 15.1.3 to prevent, detect, or investigate security incidents and manage the security of Harvey's platform and Service; 15.1.4 for business analytics and product strategy; and 15.1.5 to develop and improve new products and services and improve the performance, functionality, safety, and security of the services.”
“Harvey hosts its cloud environment in Microsoft Azure. For customers with requirements or preferences for data localization we offer processing in the EU and Switzerland or Australia. This applies to Harvey subprocessors as well.”
!AI evaluation methodology and model-change notification not disclosedGap
Testing and evaluation is marked partial: Harvey references adversarial testing and relies on citation grounding and source verification as controls against hallucination, but no evaluation methodology, benchmark results or accuracy metrics appear in the collected documents. Change management is also partial: subprocessor additions carry 30-day notice and objection rights and the subprocessor list is versioned, but no stated process exists for notifying customers when underlying models change (model swaps are announced via blog, e.g. Fable 5.1). The breach notification timeframe lives in the Security Addendum, which is public but was not in the collected set, so that is a collection limitation rather than a vendor gap.
Legal work product depends on output accuracy; a buyer should be able to see how Harvey measures hallucination and citation accuracy and be told when the model behind a workflow changes, since that can shift output quality without any visible product change.
Question for vendor: What evaluation suite and acceptance thresholds are applied before a new or updated model is released to customers, and how are customers notified when the model serving a given feature changes?
“These independent assessments reflect our ongoing investment in rigorous controls, adversarial testing, and strong assurance practices across our platform.”
“Harvey’s Assistant emphasizes cited answers and source verification, while Harvey’s agentic search writes and iteratively refines targeted searches over your sources, expanding context until it has enough information to provide a cited answer.”
“If You subscribe to receive email notifications at the Subprocessor List, then 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. You may, within 15 days of receiving the notice of the change, reasonably object to Harvey’s use of a Subprocessor on reasonable grounds relating to the protection of Customer Personal Data”
“Previous Versions 2026 April 10, 2026 Subprocessor List 2025 December 26, 2025 Subprocessor List August 11, 2025 Subprocessor List”
“Fable 5.1, Now Live in Harvey 9/1”
“Harvey will notify You in accordance with the timeframe set out under the heading “Incident Detection and Response” in the Security Addendum which is incorporated into this DPA. Harvey will provide You with all information about the Personal Data Breach as required by Data Protection Law”
?Technical dependency observed: IntercomObservation
Technical evidence (DNS, certificate, or HTTP) shows a dependency on Intercom as a application builder. This dependency is not identified in the vendor's published materials.
An undisclosed provider dependency is exactly the gap independent verification exists to surface.
Question for vendor: Verification required — Intercom appears to be involved as a application builder: confirm whether this dependency exists, and whether it processes customer data.
“Fable 5.1, Now Live in Harvey 9/1”
“*Customers that have selected Harvey processing locations outside the listed locations for Anthropic and Mistral may still access models from such AI providers as available through Harvey’s model hosting platforms (e.g., AWS Bedrock, Google Vertex).”
“Harvey contractually prohibits model providers from training on customer data and only uses your data for processing your requests, not for model improvement. Harvey requires Zero Data Retention (ZDR) by model providers.”
“No retention of any customer data: All data is processed ephemerally and then deleted. No human eyes on customer data: We call this “eyes off,” and it means that no human has access to or reviews our customer data. No AI training on customer data: Our customers’ data belongs to them and may not be used to train AI models.”
“Microsoft Corp AI provider and cloud services provider US, EU, Switzerland, Australia OpenAI, LLC AI provider US, EU, Switzerland, Australia Google Cloud Platform AI provider US, EU, Switzerland, Australia Amazon Web Services AI provider and cloud services provider US, EU, Switzerland, Australia Anthropic, PBC AI provider US* ElevenLabs Inc. AI provider (including voice services) US, EU DeepL SE AI provider (machine translation) US, EU Mistral AI SAS* AI provider EU Baseten Labs, Inc. AI provider US, EU, Australia Fireworks.ai, Inc. AI provider US, EU, Australia”
“SuSea, Inc (you.com) Web browsing US, EU Parallel Web Systems Inc. Knowledge Sources and Web browsing US, EU RELX Inc., d.b.a. LexisNexis Ask LexisNexis US”
“enter into contractual arrangements with each Subprocessor binding them to data protection obligations that are materially equivalent to, and no less protective than, those imposed on Harvey under this DPA, and to provide the same level of security provided for in this DPA. Harvey will remain fully liable to You for the performance of each Subprocessor”
?Technical dependency observed: AtlassianObservation
Technical evidence (DNS, certificate, or HTTP) shows a dependency on Atlassian as a service provider. This dependency is not identified in the vendor's published materials.
An undisclosed provider dependency is exactly the gap independent verification exists to surface.
Question for vendor: Verification required — Atlassian appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data.
“Fable 5.1, Now Live in Harvey 9/1”
“*Customers that have selected Harvey processing locations outside the listed locations for Anthropic and Mistral may still access models from such AI providers as available through Harvey’s model hosting platforms (e.g., AWS Bedrock, Google Vertex).”
“Harvey contractually prohibits model providers from training on customer data and only uses your data for processing your requests, not for model improvement. Harvey requires Zero Data Retention (ZDR) by model providers.”
“No retention of any customer data: All data is processed ephemerally and then deleted. No human eyes on customer data: We call this “eyes off,” and it means that no human has access to or reviews our customer data. No AI training on customer data: Our customers’ data belongs to them and may not be used to train AI models.”
“Microsoft Corp AI provider and cloud services provider US, EU, Switzerland, Australia OpenAI, LLC AI provider US, EU, Switzerland, Australia Google Cloud Platform AI provider US, EU, Switzerland, Australia Amazon Web Services AI provider and cloud services provider US, EU, Switzerland, Australia Anthropic, PBC AI provider US* ElevenLabs Inc. AI provider (including voice services) US, EU DeepL SE AI provider (machine translation) US, EU Mistral AI SAS* AI provider EU Baseten Labs, Inc. AI provider US, EU, Australia Fireworks.ai, Inc. AI provider US, EU, Australia”
“SuSea, Inc (you.com) Web browsing US, EU Parallel Web Systems Inc. Knowledge Sources and Web browsing US, EU RELX Inc., d.b.a. LexisNexis Ask LexisNexis US”
“enter into contractual arrangements with each Subprocessor binding them to data protection obligations that are materially equivalent to, and no less protective than, those imposed on Harvey under this DPA, and to provide the same level of security provided for in this DPA. Harvey will remain fully liable to You for the performance of each Subprocessor”
?Technical dependency observed: MicrosoftObservation
Technical evidence (DNS, certificate, or HTTP) shows a dependency on Microsoft as a service provider. This dependency is not identified in the vendor's published materials.
An undisclosed provider dependency is exactly the gap independent verification exists to surface.
Question for vendor: Verification required — Microsoft appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data.
“Fable 5.1, Now Live in Harvey 9/1”
“*Customers that have selected Harvey processing locations outside the listed locations for Anthropic and Mistral may still access models from such AI providers as available through Harvey’s model hosting platforms (e.g., AWS Bedrock, Google Vertex).”
“Harvey contractually prohibits model providers from training on customer data and only uses your data for processing your requests, not for model improvement. Harvey requires Zero Data Retention (ZDR) by model providers.”
“No retention of any customer data: All data is processed ephemerally and then deleted. No human eyes on customer data: We call this “eyes off,” and it means that no human has access to or reviews our customer data. No AI training on customer data: Our customers’ data belongs to them and may not be used to train AI models.”
“Microsoft Corp AI provider and cloud services provider US, EU, Switzerland, Australia OpenAI, LLC AI provider US, EU, Switzerland, Australia Google Cloud Platform AI provider US, EU, Switzerland, Australia Amazon Web Services AI provider and cloud services provider US, EU, Switzerland, Australia Anthropic, PBC AI provider US* ElevenLabs Inc. AI provider (including voice services) US, EU DeepL SE AI provider (machine translation) US, EU Mistral AI SAS* AI provider EU Baseten Labs, Inc. AI provider US, EU, Australia Fireworks.ai, Inc. AI provider US, EU, Australia”
“SuSea, Inc (you.com) Web browsing US, EU Parallel Web Systems Inc. Knowledge Sources and Web browsing US, EU RELX Inc., d.b.a. LexisNexis Ask LexisNexis US”
“enter into contractual arrangements with each Subprocessor binding them to data protection obligations that are materially equivalent to, and no less protective than, those imposed on Harvey under this DPA, and to provide the same level of security provided for in this DPA. Harvey will remain fully liable to You for the performance of each Subprocessor”
Assurance dimensionsweighted components of one score, not eight separate ratings
Organisation & AI governance15% of the score65
Who owns AI risk: policies, responsible-AI principles, human oversight and administrative controls.
Assessed against ISO 42001 · NIST AI RMF · AIUC-1
“Dr. Felix Wittern, Fieldfisher, whose principal office is at Amerigo-Vespucci-Platz 1, 20457 Hamburg, Germany; [email protected]”
“Harvey's in-house security team spans infrastructure, product, and operations, delivering 24/7 monitoring alongside enterprise-grade controls, including SAML SSO, audit logs, IP allow-listing, and data lifecycle management.”
“AI AI Governance AI Monitoring AI Overview View more”
“During the due diligence process, our product teams could advocate for the privacy and security requirements because they know the non-negotiables: No training on customer data, zero-day data retention, and no human review .”
Governance & accountability: vendor-evidenced, not yet independently corroborated.
AI system10% of the score47
What AI the vendor actually runs and where it sits in the product, the map everything else is judged against.
Assessed against ISO 42001 · NIST AI RMF
“Harvey is a SaaS AI productivity platform that helps lawyers and other professionals quickly produce complex, high-quality work product through a natural-language interface.”
“Harvey Assistant supports Q&A-style chat, deep analysis, and drafting. Lawyers can ask questions, analyze documents, generate memos, and produce polished drafts grounded in trusted sources and available context. Harvey Knowledge helps teams ground answers in institutional knowledge and 600+ legal databases and curated public sources across jurisdictions.”
“Introducing Harvey Tenet, our first post-trained open-weight model.”
“We use publicly available information (for example, judgments, decisions, public filings, etc.,) to develop, train, and improve our AI platform”
How AI features are evaluated before and after release, evals, red-teaming, monitoring.
Assessed against NIST GenAI · NIST AI RMF · ISO 42001
“These independent assessments reflect our ongoing investment in rigorous controls, adversarial testing, and strong assurance practices across our platform.”
“Harvey’s Assistant emphasizes cited answers and source verification, while Harvey’s agentic search writes and iteratively refines targeted searches over your sources, expanding context until it has enough information to provide a cited answer.”
How changes to models and AI features are controlled and communicated: versioning, notice, rollback.
Assessed against ISO 27001 · ISO 42001
“If You subscribe to receive email notifications at the Subprocessor List, then 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. You may, within 15 days of receiving the notice of the change, reasonably object to Harvey’s use of a Subprocessor on reasonable grounds relating to the protection of Customer Personal Data”
“In the event that new legislation and regulations are implemented that specifically govern the use of artificial intelligence solutions, both parties agree to review this DPA to ensure compliance with such legislation and regulations.”
“Previous Versions 2026 April 10, 2026 Subprocessor List 2025 December 26, 2025 Subprocessor List August 11, 2025 Subprocessor List”
AI system description: vendor-evidenced, not yet independently corroborated.
Model10% of the score75
Which model providers and versions process customer data, and how changes to them are disclosed.
Assessed against NIST GenAI · ISO 42001
“Fable 5.1, Now Live in Harvey 9/1”
“*Customers that have selected Harvey processing locations outside the listed locations for Anthropic and Mistral may still access models from such AI providers as available through Harvey’s model hosting platforms (e.g., AWS Bedrock, Google Vertex).”
“Harvey contractually prohibits model providers from training on customer data and only uses your data for processing your requests, not for model improvement. Harvey requires Zero Data Retention (ZDR) by model providers.”
“No retention of any customer data: All data is processed ephemerally and then deleted. No human eyes on customer data: We call this “eyes off,” and it means that no human has access to or reviews our customer data. No AI training on customer data: Our customers’ data belongs to them and may not be used to train AI models.”
“Microsoft Corp AI provider and cloud services provider US, EU, Switzerland, Australia OpenAI, LLC AI provider US, EU, Switzerland, Australia Google Cloud Platform AI provider US, EU, Switzerland, Australia Amazon Web Services AI provider and cloud services provider US, EU, Switzerland, Australia Anthropic, PBC AI provider US* ElevenLabs Inc. AI provider (including voice services) US, EU DeepL SE AI provider (machine translation) US, EU Mistral AI SAS* AI provider EU Baseten Labs, Inc. AI provider US, EU, Australia Fireworks.ai, Inc. AI provider US, EU, Australia”
Agent10% of the score35
No independently corroborated agent-specific assurance (AIUC-class or equivalent).
Customer data15% of the score82
Whether customer data trains models, how long it is retained, where it is processed, and how it is protected.
Assessed against ISO 27018 · OAIC APPs · EU AI Act
“Harvey contractually guarantees through our Platform Agreement that your data stays yours. We don’t use inputs, outputs, or uploaded documents to train underlying models.”
“Yes, but only if you explicitly request it. In that case, a bespoke model is created exclusively for you, and your data is never used to train models used by other customers.”
“Customers define their own data retention periods, with options as short as three hours, allowing them to tailor data handling to their specific risk profiles and internal governance requirements. Beyond retention settings, users can delete their data within the app at any time.”
“Within 30 days of either (i) termination of the Service; or (ii) upon Your reasonable request; Harvey shall, and shall direct each Subprocessor to, return to You or delete the Customer Personal Data, unless Harvey is required by law to retain such Customer Personal Data.”
“Harvey hosts its cloud environment in Microsoft Azure. For customers with requirements or preferences for data localization we offer processing in the EU and Switzerland or Australia. This applies to Harvey subprocessors as well.”
“You acknowledge that, unless You and Harvey have agreed, in Your currently operative order form or otherwise in writing, to Process and store Customer Personal Data exclusively in a different geographic location, You may transfer Personal Data to Harvey in the United States, in order for Harvey to provide the Service and for the purposes set out in this DPA.”
“When Harvey processes Personal Data contained in Account Information and Usage Personal Data, Harvey is acting as a Data Controller, and Harvey will: 15.1 Process such Personal Data, in compliance with applicable Data Protection Law, and strictly to: 15.1.1 manage Your account, and Harvey’s relationship with You, including identity verification, and providing support, required to access or use the Service; 15.1.2 to carry out Harvey’s core business operations, such as accounting, and auditing; 15.1.3 to prevent, detect, or investigate security incidents and manage the security of Harvey's platform and Service; 15.1.4 for business analytics and product strategy; and 15.1.5 to develop and improve new products and services and improve the performance, functionality, safety, and security of the services.”
Customer data treatment: the certifications above are independently corroborated, but they cover the management system — these specific disclosures are the vendor's own statements, not yet independently verified.
AI supply chain10% of the score75
The named chain of AI and infrastructure providers behind the product, and how it is disclosed.
Assessed against ISO 27001 · ISO 27017
“Microsoft Corp AI provider and cloud services provider US, EU, Switzerland, Australia OpenAI, LLC AI provider US, EU, Switzerland, Australia Google Cloud Platform AI provider US, EU, Switzerland, Australia Amazon Web Services AI provider and cloud services provider US, EU, Switzerland, Australia Anthropic, PBC AI provider US* ElevenLabs Inc. AI provider (including voice services) US, EU DeepL SE AI provider (machine translation) US, EU Mistral AI SAS* AI provider EU Baseten Labs, Inc. AI provider US, EU, Australia Fireworks.ai, Inc. AI provider US, EU, Australia”
“SuSea, Inc (you.com) Web browsing US, EU Parallel Web Systems Inc. Knowledge Sources and Web browsing US, EU RELX Inc., d.b.a. LexisNexis Ask LexisNexis US”
“enter into contractual arrangements with each Subprocessor binding them to data protection obligations that are materially equivalent to, and no less protective than, those imposed on Harvey under this DPA, and to provide the same level of security provided for in this DPA. Harvey will remain fully liable to You for the performance of each Subprocessor”
“If You subscribe to receive email notifications at the Subprocessor List, then 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. You may, within 15 days of receiving the notice of the change, reasonably object to Harvey’s use of a Subprocessor on reasonable grounds relating to the protection of Customer Personal Data”
Security foundation10% of the score74
How security problems are found, reported and handled: disclosure channels, bug bounty, incident response.
Assessed against ISO 27001 · NIST AI RMF
“Harvey conducts automated vulnerability scans, annual third-party penetration tests, and maintains continuous security monitoring.”
“# Vulnerability Disclosure Program: Contact: mailto:[email protected]”
“Harvey will notify You in accordance with the timeframe set out under the heading “Incident Detection and Response” in the Security Addendum which is incorporated into this DPA. Harvey will provide You with all information about the Personal Data Breach as required by Data Protection Law”
“Harvey’s Security Addendum includes binding terms on data protection, data access, incident response SLAs, and other controls aligned with SOC 2, ISO, GDPR and other standards. Our commitments are auditable, enforceable, and built to exceed standard vendor terms.”
Capped at 74: the evidence here is the vendor’s own account, with nothing independently corroborating it.
Vulnerability & incident handling: vendor-evidenced, not yet independently corroborated.
Independent assurance evidence10% of the score55
Whether claimed certifications and attestations are independently corroborated, checked against official registries and issuing bodies. Which certifications, and what each one does and does not establish, is stated per certificate.
Assessed against ISO 17021-1 · ISO 42006 · ISO 27006-1 · ISO 27008
“We are pleased to announce that Harvey has recently achieved AIUC-1 certification, ISO 42001 certification, and IRAP attestation, reinforcing our commitment to security, safety, reliability, and responsible AI governance.”
“Announcing Successful Completion of 2026 SOC 2 Type II Audit and ISO 27001 Certification We are pleased to announce that Harvey has successfully completed another SOC 2 Type II audit and renewed our ISO 27001 certification.”
“Harvey partners with top-tier security firms, including Schellman, NCC Group and Bishop Fox, to perform in-depth audits. This offers external validation that Harvey meets the highest standards of resilience.”
“Compliance AIUC-1 CCPA EU AI Act GDPR IRAP ISO/IEC 27001 ISO/IEC 27001 SoA ISO/IEC 27701 ISO/IEC 27701 SoA ISO/IEC 42001:2023 ISO/IEC 42001 SoA SOC 2 Type 2 TX-RAMP VPAT WCAG”
“Harvey has certified to the U.S. Department of Commerce that it adheres to the EU-U.S. Data Privacy Framework Principles (“ EU-U.S. DPF Principles” ) with regards to the processing of personal data received from the European Union (“ EU ”) and UK in reliance on the EU-U.S. DPF and the UK Extension.”
Read from the registry record above, cited, not reproduced.
None of the corroborated certifications is AI-specific (ISO/IEC 42001 or AIUC-class) — this assurance is security and data-transfer attestation, not AI-management-system assurance.
Legal & contractual10% of the score60
The contractual backbone a buyer can rely on, terms, DPAs, privacy commitments and regulatory posture.
Assessed against EU AI Act · OAIC APPs
“Platform Agreement Read Now → Addenda Service Terms Read Now → Support and Service Level Terms Read Now → Security Addendum Read Now → Acceptable Use Policy Read Now → Data Transfer Addendum Read Now → Data Processing Addendum Read Now →”
“Harvey’s Security Addendum includes binding terms on data protection, data access, incident response SLAs, and other controls aligned with SOC 2, ISO, GDPR and other standards. Our commitments are auditable, enforceable, and built to exceed standard vendor terms.”
“This Privacy Policy does not apply to any input or output generated on our online platform, or documents uploaded to our platform. We process that data on behalf of Customers and we call it (“ Customer Data” and “Content ”). Harvey’s use of Customer Data and Content received through the Services is governed by the relevant Customer Agreement. Harvey processes Customer Data and Content received through the Services as a Data Processor”
“information, assessments, or audits, to the extent required by Data Protection Law, and as necessary to confirm that Harvey is Processing Customer Personal Data in a manner consistent with this DPA. All audits and assessments will be performed in the manner set out under the heading “Customer Audit Rights” in the Security Addendum”
Legal & contractual transparency: vendor-evidenced, not yet independently corroborated.
What would strengthen assuranceeach figure is from today's score, resolving one changes the others
Resolving an evidence gap strengthens the dimension it belongs to, and with it the score: 64 → up to 84 with every identified gap resolved.
AI supply chainclick a node to focus it · drag to pan · zoom with the controls
View as list
What you inheritTrustyCyber's AI Shared Responsibility Matrix v1.0
This scan names the vendor’s AI supply chain without assessing the responsibility split, that depends on how you consume the product. Under the common managed models, a deployer inherits 16 of the matrix’s 41 responsibilities from providers, each with named evidence owed.
Map your inherited responsibilitiesAssurance evidence: certifications 9, registry checks and verification ladders, click to view
Annual audit by Schellman per Harvey; 2026 report announced. Report gated in Trust Center.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Renewed 2026 per Harvey, auditor Schellman; Statement of Applicability listed in Trust Center.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Certification announced with SoA available in Trust Center; issuer and scope not stated in collected documents.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Certification body stated as Schellman; scope and date not in collected documents.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Described as an attestation (IRAP is an assessment, not a certification); classification level and assessed region not stated.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Appears only as a Trust Center compliance item with SoA; no issuer, scope or date in collected documents.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Listed in Trust Center compliance inventory only; level and date not stated.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Self-certification to U.S. Department of Commerce stated in Privacy Policy; applies to personal data Harvey handles as controller.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Checked against Data Privacy Framework (dataprivacyframework.gov), Oct 21, 2026: Verified on the registry
Sources 13, click to view
Appendix: standards landscape 8, click to view
Where each standard behind this assessment stands today, from TrustyCyber’s continuously maintained standards radar.
Want to go further?
This scan assesses Harvey at an organisational level. Assurance can vary significantly between products, models and agents, enter the product you’re evaluating to run a deeper AI System Assurance Scan.
Scan a specific product →This automated result is based on publicly available information at the time of scanning. It is not an audit, certification, legal opinion or assurance engagement. Publicly unavailable evidence may materially change the conclusion.
Monitor for changes
Get an email if its TRUSTYCYBER Score, grade or certifications change. One credit per refresh; reading the report always stays free.