Claude Code, GDPR and client data: what to check before you start

If your website or your clients' websites hold personal data of people in the EU or UK, GDPR applies to anything you do with that data, including sending it to Claude Code. The good news is that most website work doesn't need personal data at all. When it does, Anthropic's commercial plans (Team, Enterprise and the API) come with a Data Processing Addendum that treats your organization as the controller and Anthropic as your processor, with EU Standard Contractual Clauses built in. Consumer plans (Pro and Max) don't come with that addendum, so they are a poor fit for anyone else's personal data.

Two limits are worth knowing up front. As of October 2026, Anthropic's own API offers only US or global processing, not an EU-only option; EU data residency is available through AWS Bedrock and Google Cloud, with Microsoft Foundry listed as "coming soon." And Anthropic's terms only cover Anthropic: every MCP server, plugin or hosting tool Claude Code talks to is another processor you have to account for. This is not legal advice. GDPR obligations depend on your role, your contracts and your data; have a data protection lawyer or your DPO confirm your setup.

Where personal data shows up in website work

GDPR defines personal data as "any information relating to an identified or identifiable natural person." On a typical business site that includes contact form submissions, newsletter lists, customer and order records in an online store, comments, user accounts, server and analytics logs with IP addresses, and staff bios.

Ask a simple question for each task: does Claude Code need to see any of this to do the job? Rewriting service pages, fixing layout, adding schema, building redirects and speeding up templates need none of it. Debugging a form, migrating a store or cleaning up a CRM might. Article 5 of GDPR makes "data minimisation" a core principle, and the controller "shall be responsible for, and be able to demonstrate compliance." In practice, the cheapest way to demonstrate it is to show that personal data never entered the session.

Controller and processor roles, in plain terms

A controller decides why and how personal data is processed. A processor handles it on the controller's behalf. Article 28 says a controller "shall use only processors providing sufficient guarantees," and that processing "shall be governed by a contract or other legal act."

Anthropic's Data Processing Addendum sets out its side: "With respect to Customer Personal Data, Customer is the controller and Anthropic is Customer's processor." Who "Customer" is depends on whose account is used, and that is where website projects get complicated:

Article 28(2) says a processor "shall not engage another processor without prior specific or general written authorisation of the controller." If your agency's tools include Claude Code, ask them to name it in the subprocessor list of their agreement with you. Our guide to contract terms for Claude Code work covers the clauses around accounts and data.

What Anthropic's DPA includes

The DPA is short enough to read, and worth reading. Points a website team usually asks about:

• Transfers. The Standard Contractual Clauses, "Module Two (controller to processor) or Module Three (processor to processor)," are "incorporated by reference and will be deemed to have been executed." Under Clause 17, the governing law is "the law of the Republic of Ireland," with Swiss law for transfers governed by Swiss data protection law.
• Subprocessors. You give "general authorization" for the subprocessors listed in Schedule 4, which points to Anthropic's public subprocessor page. Anthropic will give "reasonable notice" of new ones, and you may object in writing within 15 days on reasonable data privacy or security grounds.
• Breach notice. Anthropic will notify you "without undue delay, but in any event within 48 hours, after becoming aware of any Security Breach."
• Assistance. Anthropic will provide reasonable help with data protection impact assessments and with data subject requests.

The Commercial Terms add that "Customer Content is Customer's Confidential Information" and that Anthropic "may not train models on Customer Content from Services." They also set the contracting entity: "Anthropic Ireland, Limited if Customer resides in the European Economic Area ("EEA"), Switzerland or UK, and Anthropic, PBC if Customer resides anywhere else."

Where the data is processed

This is the area most often overstated. Anthropic's data residency documentation describes two settings for the Claude API. Inference geo can be "global" (the default, "Inference may run in any available geography") or "us." Workspace geo, which controls where data is stored at rest, has "us" as the only option and "can't be changed afterward." US-only inference on recent models is billed at 1.1 times the standard rate.

Anthropic's regional compliance page says you can "Choose where your data is stored and processed" and lists Europe as included on AWS Bedrock and GCP Vertex, with Microsoft Foundry "Coming soon." So, as of October 2026:

• If you need EU-only processing, the documented path is Claude through Bedrock or Vertex. Claude Code supports both as providers, but confirm with the provider which regions serve the models you plan to use, and note that their own terms then apply.
• Claude Code features that need a claude.ai account, such as cloud sessions, routines and the Chrome extension, aren't available through cloud provider credentials alone.
• Through Anthropic directly, plan on US or global processing protected by the SCCs, and confirm with counsel that this meets your obligations.

Retention, deletion and local copies

Under commercial terms, Anthropic's standard retention for Claude Code data is 30 days. Zero data retention is offered to qualified Claude for Enterprise accounts, and as a separate arrangement for Console API keys, only after Anthropic confirms eligibility. On Enterprise with the Compliance API turned on, captured Claude Code sessions are kept for 6 years by default unless your organization sets its own retention period. Two more things live outside those numbers. First, Claude Code keeps transcripts in plaintext on each developer's computer for 30 days by default, adjustable with cleanupPeriodDays. Second, anything sent with /feedback is kept for 5 years. Both matter if you have to answer an erasure request or describe your retention in a privacy notice. The full breakdown by plan is in our guide to what happens to your code and data.

Third-party tools are separate processors

Claude Code often works alongside other services: an MCP server for your CMS or CRM, a host's CLI, an analytics connector, a browser extension. Each one that receives personal data is a separate data flow with its own terms. Anthropic's zero data retention page says plainly that data processed by "third-party tools, MCP servers, or other external integrations is not covered." Before connecting a tool that can read customer records, check that vendor's DPA and transfer mechanism, and prefer read-only, narrowly scoped access. Admins can restrict which MCP servers people may connect at all, covered in managed settings and admin controls.

A working checklist for agencies and site owners

1. Classify the task. Content, design and code work: no personal data. Data work: plan it separately.
2. Use commercial terms for any work involving another person's or client's data. Never a personal Pro or Max plan.
3. Work on anonymized copies. Export a staging database with names, emails, phone numbers and addresses replaced. Debug forms with test submissions. On WordPress, our guide to WordPress staging sites covers making the copy.
4. Block access to real data files. Add permission deny rules for export folders and backups, and keep credentials out of reach as described in keeping API keys and passwords away from Claude Code.
5. Record Claude Code in your records of processing and, for agencies, in the subprocessor list you give clients.
6. Decide on residency before the project: global, US, or EU via a cloud provider.
7. Set local retention and include developer machines in your deletion process.
8. Review output before it ships. AI-written code that handles personal data, such as forms, cookie consent or account pages, needs a person's review. The working rules are in whether Claude Code is safe for a live website.

If you are hiring outside help, our guide to choosing a Claude Code agency lists questions to ask, and data handling should be one of them.

Conclusion

GDPR doesn't stop you from using Claude Code on a website, but it does make account choice and data minimization matter. Use commercial terms so Anthropic's DPA and SCCs apply, keep real personal data out of sessions wherever the task allows, know that EU-only processing currently means going through a cloud provider, and treat every connected tool as another processor. Put the roles and subprocessors in writing and have counsel confirm the setup. If you want help designing data rules and settings for a team or an agency, that is part of our Claude Code consulting. More guides are in our Claude Code Guide.

Frequently asked questions