Cognition
cognition.ai
Automated, point-in-time. Not independently reviewed or approved by TRUSTYCYBER unless expressly stated.
- Vendor discloses an unfavourable answer: Will they train on your data?
- Training on customer content: Privacy Policy permits it, DPA defers to the unpublished Agreement
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 DPA states Cognition's subprocessors are contractually or technically prevented from using Customer Data for model training, but Cognition's own AI/ML restriction is deferred to whatever the customer's Agreement says, which is not published.
What to ask for: Resolve this directly with the vendor before proceeding — this is a confirmed disclosure, not a gap to fill in.
“Licensor shall apply the same level of AI and machine learning restrictions as set forth in the Agreement. Licensor’s subprocessors are subject to equivalent restrictions through contractual obligations or technical configurations that prevent Customer Data from being used for model training purposes.”
“To customize your experience with our Services and otherwise improve our Services including, depending on the terms that apply to your use of the Services, using User Content to train, fine tune and improve the models that power our Services.”
Why it matters: The Privacy Policy, which covers Devin and Windsurf for both individuals and enterprises, lists using User Content to 'train, fine tune and improve the models that power our Services' as a processing purpose, qualified only by 'depending on the terms that apply to your use'. The DPA's AI/ML clause binds subprocessors not to train on Customer Data but states Cognition itself will apply 'the same level of AI and machine learning restrictions as set forth in the Agreement', and that Agreement is not public. The purpose-limitation clause in DPA s.3 points the other way but does not mention training explicitly.
What to ask for: Please state in writing, for our contract tier, whether any Customer Data (prompts, repository contents, outputs, feedback) is used to train, fine-tune or evaluate Cognition's own models (including the SWE family), and whether this is the default for all paid plans or requires an enterprise agreement.
“To customize your experience with our Services and otherwise improve our Services including, depending on the terms that apply to your use of the Services, using User Content to train, fine tune and improve the models that power our Services.”
“Licensor shall apply the same level of AI and machine learning restrictions as set forth in the Agreement. Licensor’s subprocessors are subject to equivalent restrictions through contractual obligations or technical configurations that prevent Customer Data from being used for model training purposes.”
“For clarity, as Data Processor, Licensor will not sell or share Customer Data, nor will Licensor use, disclose, retain, or otherwise process Customer Data (i) for a purpose other than the specific purpose of providing the Services; (ii) outside of the direct business relationship between Licensor and Customer and the written instructions received from Customer; and (iii) in a manner inconsistent with applicable Data Protection Laws.”
“Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort levels, pushing the Pareto frontier of capability and inference cost.”
Buyer questionsanswered only from the public evidence scanned, click a row for the answer
✗Will they train on your data?Concern / gap
The DPA states Cognition's subprocessors are contractually or technically prevented from using Customer Data for model training, but Cognition's own AI/ML restriction is deferred to whatever the customer's Agreement says, which is not published.
Requires written confirmation, see Before you sign ↓
“Licensor shall apply the same level of AI and machine learning restrictions as set forth in the Agreement. Licensor’s subprocessors are subject to equivalent restrictions through contractual obligations or technical configurations that prevent Customer Data from being used for model training purposes.”
“To customize your experience with our Services and otherwise improve our Services including, depending on the terms that apply to your use of the Services, using User Content to train, fine tune and improve the models that power our Services.”
✓How long do they keep your data?Clear
Customer Data under Cognition's control is permanently deleted within 90 days of a customer request or of termination/expiry of the Agreement.
“Customer Data under Licensor's control is permanently deleted within ninety (90) days: (a) upon Customer's request at any time; and (b) following any termination or expiration of the Agreement.”
“Upon termination of this DPA, Licensor will direct each Subprocessor to delete Customer Data within thirty (30) days of the termination, unless prohibited by law.”
✓Who else can access your data?Clear
The DPA's subprocessor appendix names OpenAI, Anthropic, Microsoft/Azure and Google/Vertex as 'core processing' recipients of personal data submitted as input to the Services, all located in the United States, with Databricks, xAI and Snowflake added only if enabled in Devin Desktop.
“The following entities to the extent Personal Data is submitted as Input to the Services. Entity Purpose Location OpenAI core processing United States Anthropic core processing United States Microsoft/Azure core processing United States Google/Vertex core processing United States Databricks * core processing United States xAI * core processing United States Snowflake * core processing United States * Only if enabled in Devin Desktop”
“Entity Purpose Location Microsoft Azure core processing United States AWS core processing, profile data, logs, and indices United States Google Cloud core processing, profile data, logs, and indices United States Auth0 profile data United States Hex system logs United States Sentry system logs United States Datadog system logs United States Zendesk support contacts United States Pylon support contacts United States Decagon Support Chatbot United States Agumbe (Cognition India) Support Workflow India”
!Where is your data processed?Ask the vendor
The Privacy Policy states personal information may be transferred, stored and processed throughout the world including the EU, UK and US, relying on SCCs and the UK addendum for European data; the DPA subprocessor list places all core processing in the United States with one support workflow in India. No data residency option is published.
Confirm in writing: Ask the vendor to state this in writing before signing.
“To facilitate our operations, Cognition, our third-party partners, and service providers may transfer, store, and process your personal information throughout the world, including the EU, UK and the US.”
!What happens in a security incident?Ask the vendor
The DPA commits Cognition to inform the customer without undue delay after any confirmed breach affecting Customer Data, whether caused by Cognition, its subprocessors or other third parties acting for it, and to assist with investigation.
Requires written confirmation, see Before you sign ↓
“Licensor will inform Customer without undue delay following any confirmed breach of security leading to the accidental or unlawful destruction, loss, alteration, unauthorized disclosure of, or unauthorized access to Customer Data processed by Licensor for Customer”
“Where required under applicable data protection laws, Licensor will notify Customer of any Security Breach for which that Customer is impacted and take appropriate measures to address the Security Breach, including measures to mitigate any adverse effects resulting from the Security Breach.”
Email to send the vendor9 items to confirm in writing
Hello Cognition team, We are assessing Cognition 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. What is your commitment to notify customers of a security incident affecting our data, including the timeframe? 3. In which countries and regions is our data processed and stored, and can processing be limited to a region we choose? 4. Can customers pin or exclude specific model providers for their tenant, and will Cognition notify customers before routing changes move their data to a different listed provider? 5. Confirm which deletion window (30 or 90 days) and which breach-notification standard governs, and provide the severity-tiered remediation timeframes referenced in Security Exhibit s.5.1.7. 6. Please share the current SOC 2 Type 2 report and ISO/IEC 27001:2022 certificate (with scope statement and certification body) under NDA, and confirm the FedRAMP authorization target date and sponsoring agency. 7. Intercom appears to be involved as a application builder: confirm whether this dependency exists, and whether it processes customer data. 8. Microsoft appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data. 9. Stripe 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
✓Published security exhibit and DPA give an unusually complete baselineStrong
Cognition publishes a detailed Security Exhibit (control/data-plane architecture, named Head of Security, annual third-party audits against SOC 2 Type II and ISO 27001, annual pen tests plus bug bounty, 90-day deletion) and a DPA whose Appendix C names every infrastructure, support and model-provider subprocessor with location. The customer-hosted Data Plane option, where Cognition states it cannot access Customer Data, is a meaningful differentiator for an agent that executes code against customer infrastructure.
A buyer can map most standard due-diligence questions to a public, contractual document rather than a questionnaire, and the model-provider list lets them trace where prompts and code go.
“Depending on your configuration and which Software Services features Customer access, Licensor may process Customer Data stored within Licensor infrastructure or the Licensor Data Plane. For clarity, Licensor cannot access Customer Data stored on the Customer Data Plane.”
“The following entities to the extent Personal Data is submitted as Input to the Services. Entity Purpose Location OpenAI core processing United States Anthropic core processing United States Microsoft/Azure core processing United States Google/Vertex core processing United States Databricks * core processing United States xAI * core processing United States Snowflake * core processing United States * Only if enabled in Devin Desktop”
“Customer Data under Licensor's control is permanently deleted within ninety (90) days: (a) upon Customer's request at any time; and (b) following any termination or expiration of the Agreement.”
“Licensor uses independent third-party auditors to assess the Licensor Security Program at least annually, as described in frameworks such as the SOC 2 Type II and ISO 27001 standards.”
“Entity Purpose Location Microsoft Azure core processing United States AWS core processing, profile data, logs, and indices United States Google Cloud core processing, profile data, logs, and indices United States Auth0 profile data United States Hex system logs United States Sentry system logs United States Datadog system logs United States Zendesk support contacts United States Pylon support contacts United States Decagon Support Chatbot United States Agumbe (Cognition India) Support Workflow India”
“Licensor conducts third-party penetration tests at least annually and a bug bounty program.”
“Licensor's Head of Security leads the Licensor Security Program and develops, reviews, and approves (together with other relevant internal stakeholders) Licensor's Security Measures.”
✗Training on customer content: Privacy Policy permits it, DPA defers to the unpublished AgreementGap
The Privacy Policy, which covers Devin and Windsurf for both individuals and enterprises, lists using User Content to 'train, fine tune and improve the models that power our Services' as a processing purpose, qualified only by 'depending on the terms that apply to your use'. The DPA's AI/ML clause binds subprocessors not to train on Customer Data but states Cognition itself will apply 'the same level of AI and machine learning restrictions as set forth in the Agreement', and that Agreement is not public. The purpose-limitation clause in DPA s.3 points the other way but does not mention training explicitly.
Whether source code, prompts and outputs are used to train Cognition's own SWE models is determined by contract tier, not by any public default. An enterprise buyer must obtain the no-training restriction in writing in the Agreement; self-serve and individual users should assume the Privacy Policy position applies.
Question for vendor: Please state in writing, for our contract tier, whether any Customer Data (prompts, repository contents, outputs, feedback) is used to train, fine-tune or evaluate Cognition's own models (including the SWE family), and whether this is the default for all paid plans or requires an enterprise agreement.
“To customize your experience with our Services and otherwise improve our Services including, depending on the terms that apply to your use of the Services, using User Content to train, fine tune and improve the models that power our Services.”
“Licensor shall apply the same level of AI and machine learning restrictions as set forth in the Agreement. Licensor’s subprocessors are subject to equivalent restrictions through contractual obligations or technical configurations that prevent Customer Data from being used for model training purposes.”
“For clarity, as Data Processor, Licensor will not sell or share Customer Data, nor will Licensor use, disclose, retain, or otherwise process Customer Data (i) for a purpose other than the specific purpose of providing the Services; (ii) outside of the direct business relationship between Licensor and Customer and the written instructions received from Customer; and (iii) in a manner inconsistent with applicable Data Protection Laws.”
“Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort levels, pushing the Pareto frontier of capability and inference cost.”
!Dynamic multi-provider model routing with no published model-change noticeGap
Devin is described as model-independent, selecting the model per task, and Devin Fusion routes dynamically between models and sidekick agents. The DPA names OpenAI, Anthropic, Microsoft/Azure and Google/Vertex (plus Databricks, xAI and Snowflake if enabled in Devin Desktop) as core-processing recipients of input data. The 15-day subprocessor notice covers new entities only; nothing published commits to notifying customers when the model serving their sessions changes among already-listed providers, nor which provider's terms (and country restrictions, per the AUP) apply to a given session.
For a buyer, the effective data recipient and the behaviour of the agent can change without a contractual trigger. Regulated customers with provider-specific approvals (e.g. an approved list that excludes xAI) need a control, not a list.
Question for vendor: Can customers pin or exclude specific model providers for their tenant, and will Cognition notify customers before routing changes move their data to a different listed provider?
“Devin is model-independent — we use the right model for each task, helping customers optimize price performance.”
“Introducing Devin Fusion: a hybrid-model harness that keeps frontier-level coding intelligence while cutting costs with sidekick agents and dynamic routing.”
“The following entities to the extent Personal Data is submitted as Input to the Services. Entity Purpose Location OpenAI core processing United States Anthropic core processing United States Microsoft/Azure core processing United States Google/Vertex core processing United States Databricks * core processing United States xAI * core processing United States Snowflake * core processing United States * Only if enabled in Devin Desktop”
“Use of third-party models may be subject to the Acceptable Use Policies of the respective third parties (including supported countries of such providers).”
“Licensor may update Appendix C from time to time by posting an updated DPA on its website and notifying Customer's designated contact via email at least fifteen (15) days prior to any new Subprocessor commencing processing of Customer Data.”
!Retention and breach-notice terms differ between the Security Exhibit and the DPAGap
The Security Exhibit deletes Customer Data within 90 days of request or termination; the DPA directs subprocessors to delete within 30 days and its SCC annex states retention is term plus 30 days (citing 'Section 12' where deletion actually sits in Section 11). On breaches, the DPA promises notice 'without undue delay' for any confirmed breach, while the Security Exhibit notifies only 'where required under applicable data protection laws'. The Exhibit's vulnerability clause refers to 'the following vulnerabilities' but publishes no severity tiers or remediation timeframes.
Which document prevails matters in an incident: the DPA states it takes precedence over the Agreement, but the Security Exhibit says compliance with it satisfies any security measures in the Agreement. Buyers should fix the 30-day window and unconditional breach notice in the signed DPA and ask for the missing remediation SLAs.
Question for vendor: Confirm which deletion window (30 or 90 days) and which breach-notification standard governs, and provide the severity-tiered remediation timeframes referenced in Security Exhibit s.5.1.7.
“Customer Data under Licensor's control is permanently deleted within ninety (90) days: (a) upon Customer's request at any time; and (b) following any termination or expiration of the Agreement.”
“Upon termination of this DPA, Licensor will direct each Subprocessor to delete Customer Data within thirty (30) days of the termination, unless prohibited by law.”
“Licensor will inform Customer without undue delay following any confirmed breach of security leading to the accidental or unlawful destruction, loss, alteration, unauthorized disclosure of, or unauthorized access to Customer Data processed by Licensor for Customer”
“Where required under applicable data protection laws, Licensor will notify Customer of any Security Breach for which that Customer is impacted and take appropriate measures to address the Security Breach, including measures to mitigate any adverse effects resulting from the Security Breach.”
“Licensor will use commercially reasonable efforts to address the following vulnerabilities, with each measured from (a) the date of availability of a compatible, vendor-supplied patch (with respect to publicly declared third party vulnerabilities); or (b) the date such vulnerability is confirmed (with respect to internal vulnerabilities).”
!Certifications are vendor-stated and NDA-gated; FedRAMP is In-Process, not AuthorizedGap
SOC 2 Type 2 and ISO/IEC 27001:2022 are listed on the Trust Center and referenced in the Security Exhibit and DPA, but the reports, bridge letter and pentest report sit behind an access request and NDA, so scope, issuing body and period could not be checked from public sources. The DPA calls SOC 2 a 'certification' (it is an attestation report). FedRAMP High is stated as 'In-Process' on the Marketplace, which is a pre-authorisation status. No AI-specific assurance (e.g. ISO/IEC 42001) is claimed, and the published evaluation work (productivity estimator, FrontierCode) is internal rather than independent.
The security certifications are credible on their face but remain vendor_claimed_only until the reports are obtained; none of them speak to the AI governance of an autonomous agent, so independent_assurance is assessed partial.
Question for vendor: Please share the current SOC 2 Type 2 report and ISO/IEC 27001:2022 certificate (with scope statement and certification body) under NDA, and confirm the FedRAMP authorization target date and sponsoring agency.
“Licensor uses independent third-party auditors to assess the Licensor Security Program at least annually, as described in frameworks such as the SOC 2 Type II and ISO 27001 standards.”
“Compliance CCPA SOC 2 Type 2 ISO/IEC 27001:2022 Documents Get access DOCUMENTS Finance REPORTS Network Diagram REPORTS Pentest Report REPORTS SOC 2 Report REPORTS SOC 2 Type 2 Bridge Letter COMPLIANCE ISO/IEC 27001:2022 COMPLIANCE SOC 2 Type 2”
“You will be sent an invite via email, and will be prompted to sign an NDA once in the portal. Once the NDA is signed, you will have access to view and download the resources in our Trust Center.”
“Cognition’s entire platform is now FedRAMP Class D (High) In-Process and listed on the FedRAMP Marketplace, giving engineering teams working in federal environments access to Devin, the autonomous cloud AI software engineer.”
“In our system, an agent reviews each completed Devin session — first classifying whether it produced useful output, then estimating how long a human engineer would have taken to produce the same work. We validated it by asking human engineers how long they would have spent on the same tasks.”
?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.
“The following entities to the extent Personal Data is submitted as Input to the Services. Entity Purpose Location OpenAI core processing United States Anthropic core processing United States Microsoft/Azure core processing United States Google/Vertex core processing United States Databricks * core processing United States xAI * core processing United States Snowflake * core processing United States * Only if enabled in Devin Desktop”
“Devin is model-independent — we use the right model for each task, helping customers optimize price performance.”
“Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort levels, pushing the Pareto frontier of capability and inference cost.”
“Introducing Devin Fusion: a hybrid-model harness that keeps frontier-level coding intelligence while cutting costs with sidekick agents and dynamic routing.”
“Use of third-party models may be subject to the Acceptable Use Policies of the respective third parties (including supported countries of such providers).”
“Entity Purpose Location Microsoft Azure core processing United States AWS core processing, profile data, logs, and indices United States Google Cloud core processing, profile data, logs, and indices United States Auth0 profile data United States Hex system logs United States Sentry system logs United States Datadog system logs United States Zendesk support contacts United States Pylon support contacts United States Decagon Support Chatbot United States Agumbe (Cognition India) Support Workflow India”
“Licensor maintains a comprehensive Third Party Risk Management program that assesses the security compliance of applicable third parties, including vendors, contractors and subprocessors, in order to appropriately measure and manage risk.”
?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.
“The following entities to the extent Personal Data is submitted as Input to the Services. Entity Purpose Location OpenAI core processing United States Anthropic core processing United States Microsoft/Azure core processing United States Google/Vertex core processing United States Databricks * core processing United States xAI * core processing United States Snowflake * core processing United States * Only if enabled in Devin Desktop”
“Devin is model-independent — we use the right model for each task, helping customers optimize price performance.”
“Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort levels, pushing the Pareto frontier of capability and inference cost.”
“Introducing Devin Fusion: a hybrid-model harness that keeps frontier-level coding intelligence while cutting costs with sidekick agents and dynamic routing.”
“Use of third-party models may be subject to the Acceptable Use Policies of the respective third parties (including supported countries of such providers).”
“Entity Purpose Location Microsoft Azure core processing United States AWS core processing, profile data, logs, and indices United States Google Cloud core processing, profile data, logs, and indices United States Auth0 profile data United States Hex system logs United States Sentry system logs United States Datadog system logs United States Zendesk support contacts United States Pylon support contacts United States Decagon Support Chatbot United States Agumbe (Cognition India) Support Workflow India”
“Licensor maintains a comprehensive Third Party Risk Management program that assesses the security compliance of applicable third parties, including vendors, contractors and subprocessors, in order to appropriately measure and manage risk.”
?Technical dependency observed: StripeObservation
Technical evidence (DNS, certificate, or HTTP) shows a dependency on Stripe 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 — Stripe appears to be involved as a service provider: confirm whether this dependency exists, and whether it processes customer data.
“The following entities to the extent Personal Data is submitted as Input to the Services. Entity Purpose Location OpenAI core processing United States Anthropic core processing United States Microsoft/Azure core processing United States Google/Vertex core processing United States Databricks * core processing United States xAI * core processing United States Snowflake * core processing United States * Only if enabled in Devin Desktop”
“Devin is model-independent — we use the right model for each task, helping customers optimize price performance.”
“Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort levels, pushing the Pareto frontier of capability and inference cost.”
“Introducing Devin Fusion: a hybrid-model harness that keeps frontier-level coding intelligence while cutting costs with sidekick agents and dynamic routing.”
“Use of third-party models may be subject to the Acceptable Use Policies of the respective third parties (including supported countries of such providers).”
“Entity Purpose Location Microsoft Azure core processing United States AWS core processing, profile data, logs, and indices United States Google Cloud core processing, profile data, logs, and indices United States Auth0 profile data United States Hex system logs United States Sentry system logs United States Datadog system logs United States Zendesk support contacts United States Pylon support contacts United States Decagon Support Chatbot United States Agumbe (Cognition India) Support Workflow India”
“Licensor maintains a comprehensive Third Party Risk Management program that assesses the security compliance of applicable third parties, including vendors, contractors and subprocessors, in order to appropriately measure and manage risk.”
Assurance dimensionsweighted components of one score, not eight separate ratings
Organisation & AI governance15% of the score40
Who owns AI risk: policies, responsible-AI principles, human oversight and administrative controls.
Assessed against ISO 42001 · NIST AI RMF · AIUC-1
“Licensor's Head of Security leads the Licensor Security Program and develops, reviews, and approves (together with other relevant internal stakeholders) Licensor's Security Measures.”
“You must not direct or configure the Services to take irreversible actions in production environments without implementing appropriate human oversight, review, and confirmation mechanisms.”
“If you discover that our model outputs are inaccurate, biased or harmful, please notify us at [email protected]”
AI system15% 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
“Our Services may take autonomous actions on your behalf, including writing and executing code, interacting with third-party systems, and making changes to your infrastructure.”
“Devin is a cloud agent platform for engineering teams. You work with it like a teammate — give it tasks, review its PRs, and let it handle your backlog.”
“Depending on your configuration and which Software Services features Customer access, Licensor may process Customer Data stored within Licensor infrastructure or the Licensor Data Plane. For clarity, Licensor cannot access Customer Data stored on the Customer Data Plane.”
“Devin is model-independent — we use the right model for each task, helping customers optimize price performance.”
How AI features are evaluated before and after release, evals, red-teaming, monitoring.
Assessed against NIST GenAI · NIST AI RMF · ISO 42001
“In our system, an agent reviews each completed Devin session — first classifying whether it produced useful output, then estimating how long a human engineer would have taken to produce the same work. We validated it by asking human engineers how long they would have spent on the same tasks.”
“Ground truth bias . Our ground truth is self-reported. In our video user interviews, users were aware they were talking to Cognition, which might bias them.”
How changes to models and AI features are controlled and communicated: versioning, notice, rollback.
Assessed against ISO 27001 · ISO 42001
“Licensor maintains a documented change management policy, reviewed at least annually.”
“Licensor may update Appendix C from time to time by posting an updated DPA on its website and notifying Customer's designated contact via email at least fifteen (15) days prior to any new Subprocessor commencing processing of Customer Data.”
“Licensor regularly tests and evaluates its Security Program, and may review and update this Security Exhibit at any time without notice, provided that such updates either make equivalent or enhance Security Measures and do not materially diminish the level of protection afforded to Customer Data by these Security Measures.”
AI system description: vendor-evidenced, not yet independently corroborated.
Model10% of the score60
Which model providers and versions process customer data, and how changes to them are disclosed.
Assessed against NIST GenAI · ISO 42001
“The following entities to the extent Personal Data is submitted as Input to the Services. Entity Purpose Location OpenAI core processing United States Anthropic core processing United States Microsoft/Azure core processing United States Google/Vertex core processing United States Databricks * core processing United States xAI * core processing United States Snowflake * core processing United States * Only if enabled in Devin Desktop”
“Devin is model-independent — we use the right model for each task, helping customers optimize price performance.”
“Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort levels, pushing the Pareto frontier of capability and inference cost.”
“Introducing Devin Fusion: a hybrid-model harness that keeps frontier-level coding intelligence while cutting costs with sidekick agents and dynamic routing.”
“Use of third-party models may be subject to the Acceptable Use Policies of the respective third parties (including supported countries of such providers).”
Model provider transparency: vendor-evidenced, not yet independently corroborated.
Customer data15% of the score40
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
“For clarity, as Data Processor, Licensor will not sell or share Customer Data, nor will Licensor use, disclose, retain, or otherwise process Customer Data (i) for a purpose other than the specific purpose of providing the Services; (ii) outside of the direct business relationship between Licensor and Customer and the written instructions received from Customer; and (iii) in a manner inconsistent with applicable Data Protection Laws.”
“Licensor shall apply the same level of AI and machine learning restrictions as set forth in the Agreement. Licensor’s subprocessors are subject to equivalent restrictions through contractual obligations or technical configurations that prevent Customer Data from being used for model training purposes.”
“To customize your experience with our Services and otherwise improve our Services including, depending on the terms that apply to your use of the Services, using User Content to train, fine tune and improve the models that power our Services.”
“Customer Data under Licensor's control is permanently deleted within ninety (90) days: (a) upon Customer's request at any time; and (b) following any termination or expiration of the Agreement.”
“Upon termination of this DPA, Licensor will direct each Subprocessor to delete Customer Data within thirty (30) days of the termination, unless prohibited by law.”
“To facilitate our operations, Cognition, our third-party partners, and service providers may transfer, store, and process your personal information throughout the world, including the EU, UK and the US.”
“Depending on your configuration and which Software Services features Customer access, Licensor may process Customer Data stored within Licensor infrastructure or the Licensor Data Plane. For clarity, Licensor cannot access Customer Data stored on the Customer Data Plane.”
AI supply chain10% of the score60
The named chain of AI and infrastructure providers behind the product, and how it is disclosed.
Assessed against ISO 27001 · ISO 27017
“Entity Purpose Location Microsoft Azure core processing United States AWS core processing, profile data, logs, and indices United States Google Cloud core processing, profile data, logs, and indices United States Auth0 profile data United States Hex system logs United States Sentry system logs United States Datadog system logs United States Zendesk support contacts United States Pylon support contacts United States Decagon Support Chatbot United States Agumbe (Cognition India) Support Workflow India”
“The following entities to the extent Personal Data is submitted as Input to the Services. Entity Purpose Location OpenAI core processing United States Anthropic core processing United States Microsoft/Azure core processing United States Google/Vertex core processing United States Databricks * core processing United States xAI * core processing United States Snowflake * core processing United States * Only if enabled in Devin Desktop”
“Licensor maintains a comprehensive Third Party Risk Management program that assesses the security compliance of applicable third parties, including vendors, contractors and subprocessors, in order to appropriately measure and manage risk.”
“Licensor may update Appendix C from time to time by posting an updated DPA on its website and notifying Customer's designated contact via email at least fifteen (15) days prior to any new Subprocessor commencing processing of Customer Data.”
Subprocessors & supply chain: vendor-evidenced, not yet independently corroborated.
Security foundation15% of the score55
How security problems are found, reported and handled: disclosure channels, bug bounty, incident response.
Assessed against ISO 27001 · NIST AI RMF
“Licensor conducts third-party penetration tests at least annually and a bug bounty program.”
“Licensor will inform Customer without undue delay following any confirmed breach of security leading to the accidental or unlawful destruction, loss, alteration, unauthorized disclosure of, or unauthorized access to Customer Data processed by Licensor for Customer”
“Where required under applicable data protection laws, Licensor will notify Customer of any Security Breach for which that Customer is impacted and take appropriate measures to address the Security Breach, including measures to mitigate any adverse effects resulting from the Security Breach.”
“Licensor will use commercially reasonable efforts to address the following vulnerabilities, with each measured from (a) the date of availability of a compatible, vendor-supplied patch (with respect to publicly declared third party vulnerabilities); or (b) the date such vulnerability is confirmed (with respect to internal vulnerabilities).”
Independent assurance evidence10% of the score41
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
“Licensor uses independent third-party auditors to assess the Licensor Security Program at least annually, as described in frameworks such as the SOC 2 Type II and ISO 27001 standards.”
“Compliance CCPA SOC 2 Type 2 ISO/IEC 27001:2022 Documents Get access DOCUMENTS Finance REPORTS Network Diagram REPORTS Pentest Report REPORTS SOC 2 Report REPORTS SOC 2 Type 2 Bridge Letter COMPLIANCE ISO/IEC 27001:2022 COMPLIANCE SOC 2 Type 2”
“You will be sent an invite via email, and will be prompted to sign an NDA once in the portal. Once the NDA is signed, you will have access to view and download the resources in our Trust Center.”
“Cognition’s entire platform is now FedRAMP Class D (High) In-Process and listed on the FedRAMP Marketplace, giving engineering teams working in federal environments access to Devin, the autonomous cloud AI software engineer.”
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
“Where Customer Data that originates in the European Economic Area is transferred to a country outside of Europe that is not subject to an adequacy decision, Licensor will do so in accordance with the standard contractual clauses adopted by the EU Commission on June 4, 2021”
“Contains API keys, access tokens, passwords, private cryptographic keys, or other security credentials. Contains restricted or sensitive personal data, such as biometric identifiers, government-issued identifiers (including SSN, TIN, or passport numbers), payment card data or credit card numbers, bank account numbers, or data constituting a special category under applicable privacy law.”
“on the condition that: (i) the Audit is required by law; (ii) where permitted by law, Licensor may first provide a summary of the results of a third-party audit or certification report”
“For users who have formerly enrolled for use of Devin Desktop (f/k/a Windsurf) under Exafunction, Inc., the terms “Cognition,” “we,” “us,” or “our” refer to Exafunction, Inc, and not Cognition AI, Inc., unless the license has been assigned to Cognition AI, Inc.”
“For clarity, as Data Processor, Licensor will not sell or share Customer Data, nor will Licensor use, disclose, retain, or otherwise process Customer Data (i) for a purpose other than the specific purpose of providing the Services; (ii) outside of the direct business relationship between Licensor and Customer and the written instructions received from Customer; and (iii) in a manner inconsistent with applicable Data Protection Laws.”
Legal & contractual transparency: vendor-evidenced, not yet independently corroborated.
Not graded: Agent, not applicable to this scan.
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: 49 → up to 72 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 4, registry checks and verification ladders, click to view
Listed on the Trust Center with report and bridge letter available under NDA; referenced in the Security Exhibit and DPA (the DPA calls it a 'certification'). Audit period, auditor and in-scope services not public.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Listed on the Trust Center and referenced in the Security Exhibit. Certificate, certification body and scope statement not public.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Stated as 'FedRAMP Class D (High) In-Process' and listed on the FedRAMP Marketplace for 'Cognition's entire platform' (July 2026). In-Process is not an Authorized status.
Not yet checked against a registry. Treat it as the vendor’s own claim and ask the vendor for the current certificate.
Not claimed in any public source scanned. This is the AI-management-system certification — in its absence, the vendor’s AI governance rests on its general security and privacy certifications.
Sources 12, 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 Cognition 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.