Is Claude GDPR Compliant? Consumer vs Commercial Terms Explained

A support engineer in Lyon is chasing a bug that only shows up for one customer. She copies forty lines of application log into Claude on her personal Pro account: the customer's email address, an order ID and, buried in a stack trace, a session token. Claude spots the bug in seconds. Nobody at the company knows the log left the building.

Whether that was a GDPR problem depends almost entirely on one thing: which of Anthropic's two sets of terms she was using.

Is Claude GDPR compliant?

As with any AI tool, GDPR applies to how your company processes personal data, not to the product. When staff use Claude for work, your company is normally the controller. Anthropic then acts as your processor if the use runs under its Commercial Terms (Claude for Work Team and Enterprise plans, the Claude API, and Claude through Amazon Bedrock or Google Cloud Vertex AI). On a Free, Pro or Max account, Anthropic deals with the individual under consumer terms, and your company has no contract covering the data at all.

So compliant use of Claude is possible. It's just not automatic, and it's not what happens when someone opens a personal account.

What changed in Anthropic's consumer terms in 2025?

On 28 August 2025, Anthropic updated its consumer terms. Until then, Anthropic's position had been that it didn't train on consumer chats by default. The update introduced a choice:

• Users of Claude Free, Pro and Max, including Claude Code used from those accounts, were asked to choose whether their chats and coding sessions may be used to train models. Existing users had until 8 October 2025 to make the choice.
• If a user allows training, Anthropic keeps that data for five years. If not, the existing 30-day retention period continues.
• The setting applies to new or resumed chats and can be changed later in privacy settings. Deleted conversations aren't used for future training.
• Services under the Commercial Terms are excluded: Claude for Work (Team and Enterprise), Claude Gov, Claude for Education, and API use, including through Amazon Bedrock and Google Cloud Vertex AI.

For a company, the practical reading is simple. If an employee uses a personal account, whether their work chats help train a model depends on a toggle you don't control and can't audit.

Have regulators acted on Claude?

As of October 2026, we're not aware of a European regulator decision against Anthropic comparable to the Italian Garante's actions on ChatGPT or DeepSeek. That's not a clean bill of health; it just means the questions are the ordinary GDPR ones. The European Data Protection Board widened its ChatGPT taskforce to cover AI enforcement in general in February 2025, so the same scrutiny of legal basis, transparency and training data applies to every large model provider operating in the EU.

Consumer Claude vs commercial Claude: what's different?

Two details matter for an EU company. First, Anthropic's DPA incorporates the EU standard contractual clauses and is accepted together with the Commercial Terms, so you don't need a separate signature to have one. Second, Anthropic's data residency documentation lists only two inference locations for the first-party API, us and global, and us is the only workspace location for data at rest. If you need EU processing for Claude models, the routes today are AWS Bedrock or Google Cloud Vertex AI in an EU region, under those platforms' terms.

What does GDPR ask of you when staff use Claude?

1. A lawful basis (Art. 6) for sending personal data to Anthropic, and an Art. 9 condition for special categories such as health data. Most debugging and drafting tasks don't need personal data at all, which is the cheapest way to meet this.
2. A processor agreement (Art. 28). Covered for commercial use by the DPA; not available for personal accounts.
3. A transfer assessment (Chapter V). With storage in the US, the DPA's SCCs carry the transfer, and you need a transfer impact assessment that considers US law. If Anthropic's entity is certified under the EU-US Data Privacy Framework at the time you assess, that can simplify things; check the current list.
4. Data minimization (Art. 5(1)(c)). A stack trace doesn't need the customer's email address to be debugged.
5. Transparency and records. Mention AI tools in privacy notices (Art. 13 and 14) and your record of processing (Art. 30) where they process customer or employee data.
6. A DPIA (Art. 35) where the use is likely to be high-risk, such as Claude agents working on customer records at scale.

How do you set up Claude for GDPR?

• Standardize on commercial terms. Give people a Team or Enterprise seat, or route internal tools through the API, so the DPA and no-training position apply.
• Decide on location deliberately. If EU processing is a requirement, use Claude through Bedrock or Vertex AI in an EU region. If US processing is acceptable, set inference_geo to us rather than global so you know where inference runs.
• Mask in applications. Point the Anthropic SDK at the Maskflare gateway and personal data and secrets are replaced with tokens before the request leaves, then restored in the response, including while streaming.
• Cover claude.ai in the browser. Browser protection and the desktop agent mask text typed into Claude on the web, whichever account is signed in.
• Protect coding agents. Maskflare's Claude Code hook blocks prompts containing secrets and masks tool input and output for Bash, WebFetch, Write, Edit and MCP calls. That session token in the log would have been stopped at the prompt.

from anthropic import Anthropic

client = Anthropic(
    base_url="https://<your-gateway>/v1/anthropic",
)

message = client.messages.create(
    model="claude-sonnet-4-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Why does this fail for ada@example.com? ..."}],
)
# Anthropic receives a token in place of ada@example.com;
# the reply shows the real address again.

The Anthropic masking guide walks through keys, streaming and restore settings step by step.

What are the limits of this approach?

• Masking reduces, it doesn't exempt. Masked data is pseudonymized and still personal data on your side, so you still need a lawful basis and records.
• Free-text descriptions can still identify people. Detection finds identifiers such as emails, IBANs and national IDs; a sentence like "the customer who complained on Friday" may still point at someone.
• The desktop agent installs the hooks for Claude Code, OpenAI Codex, Cursor, GitHub Copilot CLI and VS Code agent mode, so developers don't have to.
• Location is set by Anthropic or your cloud platform, not by Maskflare. Maskflare is EU-hosted, but the masked request still goes to wherever Claude runs.

What about the EU AI Act?

The EU AI Act adds duties on top of GDPR without replacing any. As a company using Claude, your main obligations are AI literacy for the staff who use it (Article 4, applying since February 2025) and transparency where people interact with an AI system you deploy, such as a customer-facing chatbot. Anthropic carries the obligations for general-purpose model providers, which have applied since August 2025. Whether any of your specific uses counts as high-risk under the Act is a question for your legal team.

This article is general information, not legal advice. Facts are as of October 2026; check Anthropic's current terms before relying on them.