Anthropic
AI API · anthropic.com · last read 2026-11-16
At a glance
- Is Anthropic GDPR compliant?
- No tool is GDPR compliant on its own: compliance depends on what you process with it and how. What follows is what Anthropic itself publishes, each answer linked to its source and the date we read it.
- Does Anthropic have a data processing agreement (DPA)?
- Yes. Anthropic publishes a data processing agreement at anthropic.com. It describes how a new subprocessor is announced but states no notice period. It commits to telling you of a personal data breach within 48 hours. Source and date
- Which version of the Anthropic DPA is current?
- On 2026-11-09, Anthropic's DPA at anthropic.com printed “Effective Feb 24, 2025”. Source and date
- Does Anthropic have a SOC 2 report?
- Anthropic says SOC 2 Type 1 and Type 2 reports are available on request. In Anthropic's words: “For more information and to request copies of our compliance documentation, please visit our Trust Portal.” We read this on 2026-10-05. Source and date
- Is Anthropic ISO 27001 certified?
- Anthropic states ISO/IEC 27001 certification (2022 edition, as Anthropic names it), and says the certificate or audit report is available on request. In Anthropic's words: “For more information and to request copies of our compliance documentation, please visit our Trust Portal.” We read this on 2026-10-05. Source and date
- Where is Anthropic's trust center?
- Anthropic publishes a page it calls “Trust Center” at trust.anthropic.com. Some of its documents sit behind a request or sign-in: the page says “Request access”. Its text appears only after JavaScript runs: a client without it, as most tooling is, gets an empty page. The page mentions SOC 2, ISO 27001, a penetration test, subprocessors, security questionnaires and a DPA. We read it on 2026-11-03. Source and date
- Does Anthropic use your data to train AI models?
- For consumer plans (Claude Free, Pro, Max (incl. Claude Code)): chats and coding sessions used to improve models only if the user chooses to allow it; exceptions: conversations flagged for safety review, explicit opt-in programs (e.g. Trusted Tester), and thumbs up/down feedback; Incognito chats never used. For business plans (Claude for Work, Claude Gov): Not used for training by default, unless the user reports feedback or bugs (e.g. thumbs up/down) or otherwise chooses to allow it. For the API (Anthropic API): Not used for training by default, unless the user reports feedback or bugs or otherwise chooses to allow it. For the API (Claude API): Retained data not used for model training without the customer's express permission. Source and date
- What is Anthropic's data retention for AI prompts and outputs?
- For business plans (Claude for Work, Claude for Enterprise, Anthropic Console): Chats and coding sessions kept until the user deletes them; then deleted from back-end within 30 days. For the API (Anthropic API): Inputs and outputs deleted within 30 days of receipt or generation, with exceptions (longer-retention services, ZDR agreements, Usage Policy enforcement, legal requirements). For the API (Claude API): Prompts and outputs not retained by default; Covered Models require 30-day retention; flagged content may be kept up to 2 years. Source and date
- Does Anthropic offer zero data retention?
- For consumer plans (Claude Free, Pro, Max, and Ultra plans): Consumer plans are not covered by ZDR. For business plans (Claude Code for Enterprise): Zero data retention by arrangement, subject to Anthropic's approval, applied per organization; User Safety classifier results still retained; exceptions for law and misuse. For business plans (Claude Code on Claude Enterprise): Separate ZDR offering for Claude Code on Claude Enterprise, when Anthropic has enabled it for the organization. For business plans (Claude Teams and Claude Enterprise product interfaces): Product interfaces not ZDR-eligible (except separate offering for Claude Code on Claude Enterprise). For the API (Anthropic API): Zero data retention by agreement. For the API (Claude API): ZDR on request via Anthropic sales, enabled per organization; not for Covered Models or stateful features. For the API (Claude Platform (Claude API)): Zero data retention by arrangement, subject to Anthropic's approval, applied per organization; User Safety classifier results still retained; exceptions for law and misuse. Source and date
- Can Anthropic process AI data in the EU?
- For the API (Claude API): No EU option: inference geo is only US or global, and data at rest (workspace geo) is only US. Source and date
- How does Anthropic transfer personal data out of the EU?
- Anthropic's agreement relies on the EU standard contractual clauses and the UK transfer addendum. We found no entry for Anthropic on the official Data Privacy Framework List. Source and date
- Where does Anthropic store EU customer data?
- The vendor states that no EU location is offered. Source and date
- Who are Anthropic's subprocessors?
- Anthropic lists 22 third-party subprocessors; 17 of them name at least one processing country outside the EEA. We last read the list on 2026-10-19. We have recorded 1 change to it, the latest on 2026-10-19. Source and date
- How do you report a security vulnerability to Anthropic?
- Anthropic publishes a security.txt at www.anthropic.com naming https://hackerone.com/4f1f16ba-10d3-4d09-9ecc-c721aad90f24/embedded_submissions/new. Source and date
The same questions for any vendor, and the rule behind each: the vendor due diligence checklist.
Identity
| Legal entity | Anthropic, PBC |
|---|---|
| HQ | US |
| Parent company | — |
| Ultimate ownership | US |
| EU contracting entity | Anthropic Ireland, Limited |
| LEI | 984500B6DEB8CEBC4Z70 |
| US CLOUD Act reach | Contracting entity incorporated in the US |
| Source | anthropic.com · 2026-10-08 |
Legal entity from the vendor's own terms. LEI, registered country and parent company from the Global LEI Index (GLEIF); ownership is left empty where GLEIF names no parent. The CLOUD Act row is arithmetic over those countries, not an assessment of risk.
Data location
| Region | EU data location | Minimum plan | Source | Observed |
|---|---|---|---|---|
| EU | No EU location | — | platform.claude.com | 2026-10-05 |
EU: The document says workspace geo, which controls where data is stored at rest, is currently available only as "us", and inference geo only as "us" or "global". No EU location is offered.
The same question for every vendor: where SaaS vendors store EU data.
Anthropic's subprocessor list names AWS, Microsoft and Google for hosting or infrastructure. Which SaaS vendors run on which cloud, and the sovereign offerings of each: sovereign clouds.
Subprocessor chain
22 third parties · 17 with a country outside the EEA · all read 2026-10-19
| Subprocessor | Country | Purpose |
|---|---|---|
| Amazon Web Services | — | Cloud Infrastructure |
| Arkose Labs | US | Fraud and abuse detection |
| Boldr | CA | User support |
| Brave Search | US | Web Search |
| Cloudflare | — | Traffic Routing (CDN) |
| ElevenLabs | US | Text to speech |
| Functional Software, dba Sentry | US | Error Handling, User Support |
| Google Cloud Platform | — | Cloud infrastructure |
| Intercom | US | User support |
| Intuition Machines, Inc. (hCaptcha) | — | Fraud and abuse detection |
| Iterable | US | Email communications |
| Microsoft Azure | — | Cloud Infrastructure |
| Nutun | ZA | User support |
| Palantir Federal Cloud Service (PFCS) | US | FedRAMP Cloud |
| Parallel Web Systems, Inc. | US | Web Search |
| Persona | US | Fraud and abuse detection, identity verification |
| Sift | US | Fraud and abuse detection |
| Stripe | US | Billing |
| TurboPuffer | US | Web Search |
| Twilio | US | Analytics, email/SMS communications |
| WorkOS | US | Security, Single Sign-On |
| Yoti | GB | Fraud and abuse detection, identity verification |
The count outside the EEA is arithmetic over the country column, not a judgement about lawfulness. A transfer outside the EEA can be perfectly lawful; whether this one is depends on your use, not on the vendor.
Every subprocessor in the registry, and which vendors name it: subprocessors.
Processing terms
From the vendor's data processing agreement · read 2026-10-09 · anthropic.com
| Term | What the agreement says |
|---|---|
| Transfer mechanism | SCC module 2 |
| Transfer mechanism | SCC module 3 |
| Transfer mechanism | Swiss FADP clauses |
| Transfer mechanism | UK Addendum |
| Notice of new subprocessors | General written authorization (SCC Clause 9 Option 2); notice period as specified in Section C.3 of the DPA |
| Notice of new subprocessors | Reasonable notice before the new subprocessor gets access to Customer Personal Data (method not stated) |
| If you object to one | Customer may object in writing within 15 days of the notice, on reasonable data privacy or security grounds; otherwise it is deemed to consent. If it objects, the parties work in good faith toward a mutually acceptable resolution |
| Obligations passed to subprocessors | contractual agreement with each Subprocessor imposing data protection obligations substantially as protective as Anthropic's obligations under this DPA; Anthropic remains liable for Subprocessors |
| Breach notification | without undue delay, in any event within 48 hours, in writing |
| Deletion when the service ends | Within 30 days of termination or expiration of the Agreement, deletes all copies of Customer Data processed by Anthropic or any Subprocessors, except where law or regulation requires storage, retention is needed to resolve a dispute between the parties, or retention is needed to combat harmful use of the Services |
| Getting your data back | If Customer requests it within 30 days of termination or expiration, returns a copy of all Customer Data in its control or possession, or provides a self-service way for Customer to do so |
| Requests from data subjects | forwards to Customer promptly any Data Subject Request it receives and may advise the data subject to submit it directly to Customer; provides reasonable and timely assistance |
| Audit | Annual external audits against industry standards. On written request, the customer gets audit reports or certificates (e.g., SOC 2), where available, subject to confidentiality. Certifications are posted at trust.anthropic.com |
| Audit | Annual independent third-party assessment reports made available to customers (request via https://trust.anthropic.com) |
| Audit | Annual independent third-party audit reports made available to customers |
| Audit | audit reports/certificates (e.g. SOC 2) on request; customer or third-party audit at customer's expense, scope mutually agreed, no more than once per 12 months unless non-compliance indicated or regulator requires |
| Transfer impact assessment | On request, Anthropic provides information reasonably necessary for Customer to complete a transfer impact assessment, to the extent required by law |
| Transfer impact assessment | on request, information reasonably necessary for Customer to complete a TIA |
| Processing on your instructions | only to provide or maintain the Services and in compliance with Customer's documented instructions (including as set out in the Agreement and this DPA), unless required by applicable law |
| Confidentiality of staff | each person authorized to process Customer Personal Data is subject to an appropriate duty of confidentiality |
| Security measures | reasonable and appropriate technical and organizational security measures, as summarized in Schedule 2 |
Terms as the agreement states them. Where it is silent on a term, there is no row; that is not a statement that the vendor does not offer it.
Which version applies
On 2026-11-09, the agreement printed: Effective Feb 24, 2025
at anthropic.com
In effect from 2025-02-24
This version is in our copies from 2026-10-09.
These terms side by side for every vendor: data processing agreements compared. What vendors give you for a transfer impact assessment: transfer impact assessments.
AI and your data
What Anthropic's own pages state about the data you send its AI products, plan by plan. Compared with other vendors on AI vendors and your data.
| Plans | Term | What the page says | Source | Observed |
|---|---|---|---|---|
| Consumer Claude Free, Pro, Max (incl. Claude Code) | Training on your data | Not used unless you opt in chats and coding sessions used to improve models only if the user chooses to allow it; exceptions: conversations flagged for safety review, explicit opt-in programs (e.g. Trusted Tester), and thumbs up/down feedback; Incognito chats never used | privacy.claude.com | 2026-10-28 |
| Consumer Claude Free, Pro, Max, and Ultra plans | Zero data retention | Not offered Consumer plans are not covered by ZDR | platform.claude.com | 2026-11-02 |
| Business Claude for Work, Claude Gov | Training on your data | Not used Not used for training by default, unless the user reports feedback or bugs (e.g. thumbs up/down) or otherwise chooses to allow it | privacy.claude.com | 2026-10-28 |
| Business Claude for Work, Claude for Enterprise, Anthropic Console | Data retention | 30 days Chats and coding sessions kept until the user deletes them; then deleted from back-end within 30 days | privacy.claude.com | 2026-10-28 |
| Business Claude Code for Enterprise | Zero data retention | On approval Zero data retention by arrangement, subject to Anthropic's approval, applied per organization; User Safety classifier results still retained; exceptions for law and misuse | privacy.claude.com | 2026-10-28 |
| Business Claude Code on Claude Enterprise | Zero data retention | On approval Separate ZDR offering for Claude Code on Claude Enterprise, when Anthropic has enabled it for the organization | platform.claude.com | 2026-11-02 |
| Business Claude Teams and Claude Enterprise product interfaces | Zero data retention | Not offered Product interfaces not ZDR-eligible (except separate offering for Claude Code on Claude Enterprise) | platform.claude.com | 2026-11-02 |
| API Anthropic API | Training on your data | Not used Not used for training by default, unless the user reports feedback or bugs or otherwise chooses to allow it | privacy.claude.com | 2026-10-28 |
| API Claude API | Training on your data | Not used unless you opt in Retained data not used for model training without the customer's express permission | platform.claude.com | 2026-11-02 |
| API Anthropic API | Data retention | 30 days Inputs and outputs deleted within 30 days of receipt or generation, with exceptions (longer-retention services, ZDR agreements, Usage Policy enforcement, legal requirements) | privacy.claude.com | 2026-10-28 |
| API Claude API | Data retention | Not stored Prompts and outputs not retained by default; Covered Models require 30-day retention; flagged content may be kept up to 2 years | platform.claude.com | 2026-11-02 |
| API Anthropic API | Zero data retention | On approval Zero data retention by agreement | privacy.claude.com | 2026-10-28 |
| API Claude API | Zero data retention | On approval ZDR on request via Anthropic sales, enabled per organization; not for Covered Models or stateful features | platform.claude.com | 2026-11-02 |
| API Claude Platform (Claude API) | Zero data retention | On approval Zero data retention by arrangement, subject to Anthropic's approval, applied per organization; User Safety classifier results still retained; exceptions for law and misuse | privacy.claude.com | 2026-10-28 |
| API Claude API | EU processing | Not in the EU No EU option: inference geo is only US or global, and data at rest (workspace geo) is only US | platform.claude.com | 2026-10-29 |
Assurance
Does Anthropic have a SOC 2 report?
Anthropic says SOC 2 Type 1 and Type 2 reports are available on request. We read this on 2026-10-05.
- Available on request “For more information and to request copies of our compliance documentation, please visit our Trust Portal.”
Source: privacy.claude.com, 2026-10-05
Of the 114 vendors in this registry whose own pages mention SOC 2, 57 say the report is available on request, 1 links to the report itself, and 56 state it without saying how to get it. The type is the one the vendor names.
Is Anthropic ISO 27001 certified?
Anthropic states ISO/IEC 27001 certification (2022 edition, as Anthropic names it), and says the certificate or audit report is available on request. We read this on 2026-10-05.
- Available on request “For more information and to request copies of our compliance documentation, please visit our Trust Portal.”
Source: privacy.claude.com, 2026-10-05
Of the 122 vendors in this registry whose own pages state ISO/IEC 27001, 16 link to the certificate, 43 say it is available on request, and 63 state it without saying how to get it. A certificate names the certification body, the scope and the expiry date: those are the three things to check.
| Scheme | Status | Scope | Source | Observed |
|---|---|---|---|---|
| HIPAA-ready configuration · HIPAA | Claimed | BAA available; article covers commercial products such as Claude for Work and the Anthropic API | privacy.claude.com | 2026-10-05 |
| ISO 27001:2022 · ISO27001 | Available on request | Information Security Management; article covers commercial products such as Claude for Work and the Anthropic API | privacy.claude.com | 2026-10-05 |
| ISO/IEC 42001:2023 · ISO42001 | Available on request | AI Management Systems; article covers commercial products such as Claude for Work and the Anthropic API | privacy.claude.com | 2026-10-05 |
| SOC 2 Type I & Type II · SOC2T2 | Available on request | article covers commercial products such as Claude for Work and the Anthropic API | privacy.claude.com | 2026-10-05 |
Claimed means the vendor asserts the scheme on its own page and offers nothing further. Available on request means the report or certificate exists but sits behind a request, a login or an NDA. Neither is a certificate we have seen. That is a distinction an auditor makes, so we make it too.
SecNumCloud, C5, ENS and HDS for every vendor, with the pages for ISO standards, SOC 2 and TISAX: certifications compared.
Official registers
| Register | Offering, as the register names it | Status | Level | Valid until | Observed |
|---|---|---|---|---|---|
| Global LEI Index (GLEIF) | ANTHROPIC, PBC · ANTHROPIC, PBC | ISSUED | US-DE | 2027-01-28 | 2026-10-08 |
| GPAI Code of Practice signatories (European Commission) | General-Purpose AI Code of Practice (AI Act), as a provider of general-purpose AI models · Anthropic | signatory | — | — | 2026-11-09 |
- ACN qualified cloud (Italy): checked 2026-11-16, no entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
- ISO 27001 on the basis of IT-Grundschutz (BSI, Germany): checked 2026-11-16, no entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
- CISPE Data Protection Code of Conduct (GDPR Art. 40): checked 2026-11-16, no entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
- DORA critical ICT third-party providers (ESAs): checked 2026-11-16, no entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
- Data Privacy Framework List (US): checked 2026-11-16, no entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
- ENS certificates of conformity (CCN, Spain): checked 2026-11-16, no entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
- EU Cloud Code of Conduct (GDPR Art. 40): checked 2026-11-16, no entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
- FedRAMP Marketplace (US): checked 2026-11-16, no entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
- GPAI Code of Practice signatories (European Commission): checked 2026-11-16, 1 entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
- SecNumCloud qualified services (ANSSI, France): checked 2026-11-16, no entry under Anthropic; Anthropic Ireland, Limited; Anthropic, PBC
What the authority publishes, not what the vendor claims. An entry covers the offering it names and nothing else: a FedRAMP authorisation for a government edition says nothing about the commercial service. No entry is not a finding about security - most vendors never apply - and an entry filed under a name not listed above would not have been found.
What an entry on the Data Privacy Framework List covers, and which vendors hold one: EU-US Data Privacy Framework.
Support window
| Product | Support period | State | SBOM | Source | Observed |
|---|---|---|---|---|---|
| Claude Code | — | Not declared | unknown | code.claude.com | 2026-10-19 |
Vulnerability reporting: hackerone.com (security.txt, expires 2026-12-31; read 2026-10-17)
How other vendors publish theirs: security.txt across the registry.
Documents
Where is Anthropic's trust center?
Anthropic publishes a page it calls “Trust Center” at trust.anthropic.com. Some of its documents sit behind a request or sign-in. Its text appears only after JavaScript runs: a client without it, as most tooling is, gets an empty page. The page mentions SOC 2, ISO 27001, a penetration test, subprocessors, security questionnaires and a DPA. We read it on 2026-11-03. Open trust.anthropic.com
- Name “Welcome to the Anthropic Trust Center”
- Request step “Request access”
Source: trust.anthropic.com, 2026-11-03. Read from the page's text, without a model; every vendor's trust center.
| Document | Disclosure | Transcription | HTTP | Observed |
|---|---|---|---|---|
| ai | Public | full | 200 | 2026-10-28 |
| ai-2 | Public | full | 200 | 2026-10-28 |
| ai-3 | Public | full | 200 | 2026-10-28 |
| ai-4 | Public | full | 200 | 2026-10-28 |
| ai-5 | Public | full | 200 | 2026-11-02 |
| ai-6 | Public | full | 200 | 2026-10-29 |
| certifications | Public | full | 200 | 2026-10-05 |
| dpa | Public | full | 200 | 2026-10-09 |
| lifecycle-2 | Public | full | 200 | 2026-10-19 |
| residency | Public | full | 200 | 2026-10-05 |
| subprocessors | Public | full | 200 | 2026-10-19 |
| tia | Public | full | 200 | 2026-10-20 |
History
| Detected | What moved | Why | Source |
|---|---|---|---|
| 2026-10-19 | Subprocessors added Intuition Machines, Inc. (hCaptcha); Parallel Web Systems, Inc. | the vendor changed it | trust.anthropic.com |
Every reviewed change across the registry, newest first: vendor changes.