Anthropic

Under the Commercial Terms of Service, Anthropic is contractually barred from training models on Customer Content (inputs and outputs) from the commercial Services.

Requires written confirmation — see Before you sign ↓

Deleted consumer conversations are removed from history immediately and purged from back-end systems within 30 days.

Requires written confirmation — see Before you sign ↓

The DPA operates a subprocessor regime with a published list (anthropic.com/subprocessors), advance notice of additions, and a 15-day customer objection window; Anthropic remains liable for subprocessor acts.

Confirm in writing: Ask the vendor to state this in writing before signing.

Personal data is processed on servers in the US and other countries outside the EEA and UK, with transfers relying on adequacy decisions and Standard Contractual Clauses; the DPA adds UK and Swiss addenda.

Confirm in writing: Ask the vendor to state this in writing before signing.

The DPA commits to written breach notification without undue delay and in any event within 48 hours of becoming aware of a Security Breach.

Confirm in writing: Ask the vendor to state this in writing before signing.

A versioned Responsible Scaling Policy with published redlines and risk reports, per-model system cards stating training data and safety-level decisions, and named external red-team partners with adverse results disclosed alongside favourable ones.

The transparency artefacts a buyer usually has to request under NDA are public here, and the willingness to publish unfavourable evaluation results is itself evidence the favourable ones are honest.

The public Commercial Terms bar training on Customer Content; the DPA adds 48-hour breach notice, 30-day post-termination deletion extending to subprocessors, a no-sell/no-share commitment, a published subprocessor list with objection rights, and customer-side IP indemnification covering training data.

These commitments sit in signable contract documents rather than rewritable marketing pages, which is the strongest form a public commitment can take short of independent verification.

The only certification evidence on the collected public surface is the DPA's statement of annual external audits with SOC 2 named as an example; the certificates themselves, and any ISO-family certifications, are distributed via the gated trust.anthropic.com portal.

Scope, issuer and validity of the audit programme cannot be assessed from public pages, so independent assurance rests on a vendor assertion until portal documents are obtained and reviewed.

Question for vendor: Provide the current SOC 2 Type II report and any ISO/IEC 27001 or 42001 certificates, including scope statements and validity dates.

Commercial Customer Content is contractually excluded from training, while consumer Claude.ai trains on inputs and outputs unless the user opts out, and safety-flagged or user-reported conversations are used even after an opt-out.

An organisation's exposure depends on which agreement its usage actually falls under: employees using personal consumer accounts for work content are outside the commercial no-training commitment.

Question for vendor: Confirm in writing which agreement governs each access path in scope, and that all organisational use falls under the Commercial Terms' no-training commitment.

Inbound, an RFC 9116 security.txt routes to a HackerOne programme under a published policy; outbound, a coordinated-disclosure policy governs vulnerabilities Anthropic's own AI tooling finds in third-party software, with 90-day defaults and 7-day targets for actively exploited criticals.

A published policy for AI-discovered vulnerabilities is rare and signals the security function has thought past its own perimeter.

The vendor's own materials name Google as the model provider, but DNS, certificate, or HTTP evidence points to Anthropic in that role instead.

A supply-chain claim that doesn't match what is technically observable is a disclosure question worth raising directly, not assuming either side is wrong.

Question for vendor: Can you confirm whether Google or Anthropic is the actual model provider?

The vendor's own materials name Microsoft Azure as the platform provider, but DNS, certificate, or HTTP evidence points to AWS in that role instead.

A supply-chain claim that doesn't match what is technically observable is a disclosure question worth raising directly, not assuming either side is wrong.

Question for vendor: Can you confirm whether Microsoft Azure or AWS is the actual platform provider?

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.

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.

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.

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.