Is Claude GDPR Compliant? Anthropic Assurance for UK Businesses

Is Claude GDPR compliant?

The short answer: Claude can be deployed in a UK GDPR compliant way, but compliance is a property of how you contract and configure it, not of the product itself. Anthropic provides the building blocks. You still have to assemble them.

Assuring Anthropic as a supplier means two separate things:

• Contractual and evidential assurance from Anthropic: the right terms, a Data Processing Addendum (DPA), and certification evidence for your vendor risk register
• Configuration that matches your commitments: residency, retention, and training settings that line up with what you and your clients have agreed

This guide works through both, written for UK businesses that operate under UK GDPR and ISO 27001. It reflects Anthropic’s published position as of June 2026. The specifics move quickly, so treat Anthropic’s Trust Centre and regional compliance page as the authoritative source when you sign each order form. If you are weighing up AI delivery risks more broadly, our guide to the risks of AI-augmented development covers the wider governance picture.

Does Claude train on your data? Why the plan you are on matters

Anthropic’s Commercial Terms prohibit training on your data; its Consumer Terms allow it by default. This single distinction is the most important control, and it is the first thing to get right.

Under the Commercial Terms (the API, and Claude for Work Team and Enterprise plans), Anthropic states it will not use your inputs or outputs to train its models. The only exception is content you explicitly submit as feedback, and Team and Enterprise administrators can disable the feedback feature entirely.

The Consumer Terms (Claude Free and Pro) work the other way. Since Anthropic’s consumer terms update in 2025, consumer conversations are used for training unless the user opts out, with extended retention for those who allow it.

The practical rule for any work touching client or personal data:

• Use the API or a Team or Enterprise plan under Commercial Terms
• Never let staff use personal Free or Pro accounts for business data
• Put an acceptable use policy in writing and enforce it

If your team uses AI coding tools day to day, our post on maintaining ISO 27001 compliance with AI development tools covers the policy and control side in more depth.

What certifications does Anthropic hold?

Anthropic holds the certifications most UK vendor assessments ask for, and publishes them through its Trust Centre at trust.anthropic.com:

• ISO/IEC 27001:2022 (information security management)
• ISO/IEC 42001:2023 (AI management systems)
• SOC 2 Type I and Type II
• HIPAA-ready configuration with a Business Associate Agreement available
• CSA STAR registration

Certification status, subprocessor lists, and penetration test summaries are available without a non-disclosure agreement (NDA). The full SOC 2 Type II report is available under NDA. Anthropic’s own certification summary lists the current set.

What the certificates do and do not cover

These certifications cover Anthropic’s infrastructure and management systems. They do not cover:

• Your application or integration built on top of Claude
• The correctness or security of any code Claude generates
• Your own data handling before a prompt reaches the API

That residual scope stays with you, and it is worth stating explicitly in your own information security management system (ISMS) so an auditor does not have to ask.

How does Anthropic’s DPA satisfy UK GDPR?

Anthropic’s Data Processing Addendum is incorporated into the Commercial Terms of Service, so accepting the Commercial Terms accepts the DPA. It gives you the three things UK GDPR requires from a processor relationship:

• Article 28 processor terms, with you as controller and Anthropic as processor
• EU Standard Contractual Clauses (Modules 2 and 3 under Commission Implementing Decision (EU) 2021/914) for EU data transfers
• The UK International Data Transfer Addendum, which is the mechanism that satisfies your UK GDPR transfer obligation for data going to Anthropic in the US

Anthropic’s privacy centre explains how to view and sign the DPA. File the executed copy, including the UK Addendum, in your ISMS document set.

Subprocessor management

Under the DPA you grant general authorisation for Anthropic to use subprocessors, and Anthropic provides 15 days’ notice of new subprocessors so you can object. For your own compliance you need to:

• Subscribe to subprocessor change notifications via the Trust Centre
• List Anthropic as a subprocessor in your Record of Processing Activities
• Flow the arrangement through to your own client DPAs

Can you get EU or UK data residency with Claude?

There is no EU-only or UK-only residency option on the first-party Anthropic API. This is the part most people get wrong, and it is the crux of any data sovereignty commitment you make to clients.

What the first-party API offers

The first-party API gives you two data residency controls:

• Inference geography: an inference_geo request parameter (or workspace default) with two values, us and global. The default is global, meaning inference may run in any available region. US-only inference is priced at 1.1x the standard rate on current models.
• Workspace geography: controls where data is stored at rest. US is currently the only option.

So the direct API gives you multi-region or US-pinned processing under the SCCs and UK Addendum. It does not give you European residency.

The hyperscaler routes that work today

If a client contract or your own policy mandates EU or UK in-region processing, the route is through a hyperscaler, not the first-party API:

• AWS Bedrock runs Claude in-region in Ireland (eu-west-1) and Stockholm (eu-north-1). Other European regions, including Frankfurt and London, use the EU cross-region inference profile, which keeps routing within the European geography. Verify the current region list for the specific model on Claude in Amazon Bedrock before committing.
• Google Vertex AI offers Claude with EU processing via Frankfurt or the EU multi-region endpoint.
• Microsoft Foundry lists EU-native support as targeted for 2026. Today, a Foundry deployment still routes inference to Anthropic-managed infrastructure regardless of the Azure region you select, so the residency guarantee does not yet apply on that path.

Routing through Bedrock or Vertex shifts the primary infrastructure processing relationship to AWS or Google, with Anthropic’s model terms sitting on top. Factor that into your Record of Processing and client DPAs.

Is Claude in Microsoft 365 Copilot inside the EU Data Boundary?

No, and this catches out organisations that rely on the EU Data Boundary for Microsoft 365 compliance. Since January 2026, Anthropic models in Microsoft 365 Copilot, Copilot Studio, and Power Platform run under Microsoft’s subprocessor arrangement with Anthropic, governed by Microsoft’s Product Terms and DPA rather than Anthropic’s.

Microsoft states that Anthropic models in these products are currently excluded from the EU Data Boundary and, where applicable, in-country processing commitments. They are disabled by default for EU, EFTA, and UK tenants. A global administrator has to opt in through the Microsoft 365 admin centre, and that decision should go through your change control with a documented risk assessment, not a casual toggle.

How long does Anthropic retain your data?

Anthropic states it automatically deletes API inputs and outputs from its backend within 30 days of receipt or generation. The exceptions matter for your DPA schedule:

• Usage Policy enforcement: flagged inputs and outputs retained up to two years, classification scores up to seven years
• Explicit feedback: retained for five years
• Legal requirements: retention where required by law

For sensitive client data, Zero Data Retention (ZDR) is available for qualifying customers. Under ZDR, data sent through eligible features is not stored after the API response is returned. ZDR applies per product and feature rather than as a blanket guarantee, so confirm the scope for each Anthropic product you use, including Claude Code. Enterprise plans also offer custom data retention controls for conversation history.

Document the retention configuration you choose, per client where it varies, in your ISMS and your client-facing DPAs.

How do you assess Anthropic under ISO 27001?

Treat Anthropic like any other critical cloud supplier under your ISO 27001 supplier controls (A.5.19 to A.5.23), with AI-specific additions. As a minimum, your supplier assessment record should hold:

• The ISO 27001:2022 and ISO 42001:2023 certificates
• The SOC 2 Type II report (requested under NDA via the Trust Centre)
• The penetration test summary
• The executed DPA with SCCs and UK Addendum
• The current subprocessor list, with change notifications subscribed

Then cover the items that are specific to an AI model provider:

• Data classification: define which categories of data may enter prompts, and enforce it. Personal data, special category data, and client confidential data each need an explicit decision.
• DPIA: run a data protection impact assessment where Claude processes personal data, and a fresh one where the processing changes materially.
• Record of Processing: add Anthropic (or AWS and Google, on hyperscaler routes) with the transfer mechanism named.
• Residual scope: record that Anthropic’s certifications cover its infrastructure, not your application or generated code.

Talk Think Do develops software under an ISO 27001:2022-certified information security management system, and this is the supplier assessment pattern we apply to our own AI tooling. Our AI readiness checklist covers the wider governance dimensions, and if you are building AI features on top of Claude, the EU AI Act guide for UK businesses covers the regulatory layer that sits above data protection.

Anthropic GDPR compliance checklist: what to do, in order

1. Get onto Commercial Terms. Confirm every business use of Claude runs through the API or a Team or Enterprise plan. Accept the DPA and file the executed copy, with the UK Addendum, in your ISMS document set.
2. Decide your residency posture per client. Where EU or UK-only processing is required, design for Bedrock or Vertex AI EU regions. Otherwise, set inference_geo deliberately rather than accepting the global default by accident.
3. Collect the certification evidence. Request the SOC 2 Type II report under NDA, save the ISO 27001 and 42001 certificates and the pen test summary, and complete a supplier assessment record against A.5.19 to A.5.23.
4. Set retention deliberately. Rely on the standard backend deletion window, agree ZDR for eligible products, or configure Enterprise retention controls, and document the choice per data category.
5. Update your records. Add Anthropic to your Record of Processing and your client-facing subprocessor lists, subscribe to subprocessor change notices, and run a DPIA where personal data is involved.
6. Re-verify before each engagement. Check the Trust Centre and the regional compliance page at the point you sign each order form. Residency options, ZDR eligibility, and the Foundry EU timeline are all moving through 2026.

If you are scoping an AI build and need the compliance posture designed in from the start, our AI development and implementation service covers exactly this, or book a consultation to talk it through.