Is Claude GDPR compliant and safe for business?

Last updated: 8 Oct 2026.

Yes, Claude can be run in a GDPR-compliant way, provided you use the right plan. On Claude's commercial terms (Team, Enterprise and the API), Anthropic does not train its models on your Customer Content, offers a Data Processing Agreement, deletes chats you remove within 30 days, and holds SOC 2 Type II, ISO 27001:2022 and ISO/IEC 42001:2023 certifications. The free and Pro consumer plans are not covered by the DPA, so businesses should not run client data through them. Facts checked 12 Sep 2026.

This post covers the specific compliance and security questions we're asked before a Claude rollout goes ahead. If you want the wider case for adopting Claude first, see our piece on Claude AI agents for business.

Does Claude train on your business data?

No, not if you're on a commercial plan. Under Anthropic's commercial terms, which cover Team, Enterprise and the API, Anthropic does not use your Customer Content (your prompts and outputs) to train its models. This is a contractual commitment written into the terms you sign, not a setting you have to remember to switch off, and it only applies once you've moved off the free consumer product.

How long Claude keeps your data

Retention periods differ by plan and by data type. The table below sets out the defaults.

From 09 Jun 2026, Anthropic also applies a specific 30-day retention policy to inputs and outputs from what it terms "Covered Models", worth checking against your own model list if you're relying on a particular retention period for a compliance sign-off.

Zero data retention (ZDR)

Zero data retention (ZDR) is a per-organisation arrangement, agreed directly through Anthropic's sales team, under which eligible API traffic and Claude Code activity are not retained at all once a response has been returned. ZDR is available, but only for API workloads, and only by arrangement. It does not extend to the standard Team or Enterprise chat interface, to the Console, or to the consumer Free and Pro plans, all of which follow the standard retention periods set out above. See Anthropic's API and data retention documentation for the current scope.

Where Claude actually processes your data, checked 12 Sep 2026

This is the question UK buyers ask most often and the one with the least comfortable answer. First-party Claude has no EU or UK inference region. Anthropic's pricing documentation on platform.claude.com (checked 12 Sep 2026) shows the API's inference_geo parameter accepting two values: "us", which pins inference to the United States and applies a 1.1x multiplier across input, output and cache pricing, and "global", which is the default and the standard price. There is no European option, and on models earlier than Claude 4.6 the parameter is not supported at all.

If in-region European processing is a hard requirement, the route is not the Claude API. Anthropic's documentation points to the partner-operated platforms: Amazon Bedrock and Google Cloud both offer regional and multi-region endpoints that guarantee routing through a specific geography, at a 10% premium over the global endpoints. Coverage varies by model and changes after launch: Claude Sonnet 5.5 launched on Bedrock with global routing only and has since gained US and EU geographic profiles, while Claude Haiku 5.5 launched with them (checked 7 Oct 2026). An EU profile routes across European regions rather than keeping processing in the UK, so check the model you plan to use. You buy Claude through AWS or Google Cloud, under their regional controls, rather than from Anthropic directly.

So for a UK organisation on Team, Enterprise or the first-party API, the compliance position rests on contract rather than geography. Anthropic's Data Processing Agreement covers international transfers with EU Standard Contractual Clauses and the UK-required transfer safeguards. That is a recognised and workable basis. It is not the same thing as your data staying in the UK or the EEA, and a DPO who asks "where does it run?" deserves the accurate answer rather than a reassuring one.

One practical note on models. The current line is Claude Fable 5.1 and the limited-availability Claude Mythos 5.1 at the top, Claude Opus 5.5 beneath them, Claude Sonnet 5.5 as the general-purpose model and Claude Haiku 5.5, released 7 Oct 2026, as the cheapest. Claude Haiku 4.5 is still available, and Claude Opus 5, Claude Sonnet 5, Claude Opus 4.8 and Claude Fable 5 remain available as legacy models. The residency parameter applies only to Claude 4.6 and later, so an older pinned model in your stack may quietly be a compliance gap as well as a capability one. If you run Opus 5 today, here is what changes when you move to Opus 5.5.

The adequacy question sitting behind all of this

In Trump v. Slaughter, No. 25-332, decided 29 Jun 2026, the US Supreme Court held that Federal Trade Commission commissioners are removable at will, overruling Humphrey's Executor. The link to EU-US data transfers is analysis rather than holding: the 2023 adequacy decision relies on the FTC's independence, noyb wrote to the European Commission the same day asking it to withdraw adequacy, and Skadden (28 Jul 2026) says the decision "could increase the risk" that European courts annul it. The framework stands today and its foundations are contested. UK firms relying on the UK Extension to the Data Privacy Framework should have Standard Contractual Clauses or the IDTA in place as a fallback rather than assume the current position holds.

If processing location is the constraint you are working to, our UK and EU data residency comparison sets out what each vendor offers at each tier, with the dates the position was checked.

Certifications and attestations

Anthropic maintains a set of independent certifications and attestations that cover Claude's commercial plans. The table below reflects the current position.

Anthropic publishes and updates the current list on its trust and compliance page, which is worth checking directly before quoting a certification in a customer-facing document.

Deploying Claude safely: what to set up before go-live

Before any personal data flows through Claude, three things need to be in place: a signed Data Processing Agreement that accurately describes the processing, confirmation from your Data Protection Officer (DPO) of the lawful basis you're relying on, and clarity on international transfers. None of this is legal advice, more a checklist your DPO will expect answered before sign-off.

International data transfers

Anthropic's infrastructure and operations span multiple jurisdictions, so UK or EU personal data processed through Claude will typically involve a cross-border transfer. As provided for in Anthropic's Data Processing Agreement, transfers are governed by the safeguards it sets out, including EU Standard Contractual Clauses together with the UK-required transfer safeguards.

Single sign-on and admin controls

Single sign-on (SSO) is available on both the Team and Enterprise plans, not on Enterprise alone. It's delivered via SAML, with support for Okta, Google and Microsoft Entra as identity providers, which lets IT de-provision access centrally the moment someone leaves rather than relying on a separate Claude password.

Claude's browser and agent access (formerly Cowork)

Updated 21 Sep 2026. On 16 Sep 2026 Anthropic merged Cowork into the main Claude product. There is no longer a separate Cowork mode to switch on: agent actions, including operating a browser and working in connected apps, are available in any conversation, and by default Claude asks before it takes each action. The rollout is staged, with Pro and Max first and Team and Free plans following. Anthropic’s wording for business customers is that “Enterprise admins will hear from us at least 30 days before anything changes for their organizations.” The governance decision has not gone away. It has moved from “do we turn Cowork on” to “which actions may Claude take, in which systems, and who may switch off ask-before-acting”. We set that out in what the Claude Cowork merge means for your AI policy.

Claude Code and software your staff build with it

The questions above cover using Claude itself. An app a colleague builds with Claude Code or another AI coding tool is new software, and it needs its own review before customers or real data touch it. Check who can see the data, whether API keys or passwords sit in the code, and whether the repository belongs to the business or to one person's account. If you are not sure of the answers, have an AI-built app reviewed before it goes live.

If staff will keep building this way, it is simpler to put the controls in first. We can set Claude Code up with company-owned accounts and security rules, so their work sits in a GitHub organisation the business owns and changes are reviewed and scanned before they are merged. When Claude Code, or a similar tool, runs as a cloud agent on the vendor's machines rather than on a laptop, the repository and any secrets it can reach go with it. Our guide to cloud coding agents and data security covers what to settle before that happens.

Do you need a DPIA?

A UK organisation processing personal data at scale through Claude, particularly special category data or data covering a large number of individuals, will likely need a Data Protection Impact Assessment (DPIA) before go-live. That's a decision for your DPO, not something SpotDev or Anthropic can rule on, but the ICO's DPIA guidance is the standard starting point for working out whether one is required and what it should cover.

Five steps to a safe Claude rollout

Once the paperwork above is under way, a Claude rollout usually follows the same five steps. For a fuller walkthrough of the mechanics, see our Claude for Work deployment guide for UK companies.

• Pick the right plan: use a business or enterprise tier so the no-training default, the DPA and the admin controls all apply.
• Sign the paperwork: get the DPA in place and have your DPO confirm the lawful basis and the transfer cover before real data flows.
• Turn on SSO and define admins: connect Claude to your identity provider and name a small, accountable group of administrators.
• Set usage rules: write a short, plain policy covering what staff may and may not put into Claude, and brief everyone on it.
• Start with a contained pilot: begin with one team and one or two use cases, review what is being logged, then widen access.

If you're building wider AI governance rather than a single rollout, our AI governance framework for mid-sized UK businesses covers policy, oversight and risk beyond this post. For our own accreditations and delivery record behind this checklist, see our credentials.

Where SpotDev fits

SpotDev specialises in Claude implementation for UK organisations. Our in-house team has completed over 300 technology projects with no subcontracting, so the people who design your security and governance setup are the same people who build it. Rather than reselling tools, we handle the engineering and configuration needed to establish controlled, audited deployments. Fixed-price packages range from £8,000 to £45,000, with typical first rollouts completed in two to three weeks, via our Claude implementation packages.

SpotDev designs, builds and deploys custom Claude agents and enterprise Claude rollouts for UK businesses, with fixed packages from £8,000 to £45,000 and a first rollout live in two to three weeks. Explore our Claude implementation packages or get in touch about a rollout.

FAQs

Is Claude GDPR compliant?

Yes, on the right plan. Anthropic's Data Processing Agreement covers the Team, Enterprise and API tiers, but not the Free or Pro consumer plans, so client or staff personal data should never be processed on those. Whole-organisation compliance also depends on your own configuration, the lawful basis you rely on, and the usage rules you set for staff.

Does Claude train on my business data?

No. Under Anthropic's commercial terms, covering Team, Enterprise and the API, your Customer Content is not used to train Anthropic's models. This is a contractual commitment written into those terms, not a setting to remember to switch off, though it only applies once you're on a commercial plan rather than the free consumer product.

How long does Claude keep deleted data?

By default, chats you delete in Team or Enterprise, and API inputs and outputs, are removed from Anthropic's back-end within 30 days. Some categories are held longer: content flagged for trust and safety review for up to 2 years, and safety classification scores for up to 7 years, under Anthropic's published retention schedule.

What is zero data retention and who can get it?

Zero data retention (ZDR) is a per-organisation arrangement agreed directly with Anthropic's sales team, covering eligible API traffic and Claude Code, where nothing is retained after the response is returned. It does not cover the standard Team or Enterprise chat interface or the Console, so most day-to-day chat use still follows the normal retention periods above.

Does Claude support single sign-on?

Yes. SSO is available on both the Team and Enterprise plans, not Enterprise only, delivered via SAML, with support for Okta, Google and Microsoft Entra as identity providers. That lets IT de-provision access centrally the moment someone leaves, rather than relying on a separate Claude password.

Do we need a DPIA before deploying Claude?

Usually, yes, if you're processing personal data at scale or handling special category data. It's a decision for your DPO, not a call SpotDev or Anthropic can make for you, but the ICO's DPIA guidance is the standard reference point for working out whether one is required and what it should cover.

Does Claude offer UK or EU data residency?

Not on the first-party Claude API, Team or Enterprise. Anthropic's pricing documentation (checked 12 Sep 2026) shows the API offering US-only inference at a 1.1x price multiplier or global routing as the default, with no European option. In-region European processing is available by running Claude through Amazon Bedrock or Google Cloud regional endpoints, at a 10% premium. Otherwise the transfer is covered by Standard Contractual Clauses and UK safeguards in the Data Processing Agreement, not by in-region processing.

Is it safe to let Claude take actions, not just answer?

Since 16 Sep 2026 agent actions are part of the main Claude product rather than a separate Cowork mode. By default Claude asks before each action, and Team and Enterprise owners control whether members may use automatic approval and cloud execution. Treat it as a governance decision rather than a feature toggle: agree which systems Claude may act in and who approves changes before the rollout reaches your organisation. Detail in our post on the Cowork merge.

Is Claude safe for business use?

Yes, provided you use the right plan and set it up properly. Team, Enterprise and API deployments come with a DPA, no training on your content by default, and independent certifications; the free consumer product does not. Safety in practice also depends on the access controls, usage rules and DPIA work covered above.