EU AI Watch: Claude Code privacy complaint exposes the AI Act’s limits

An email address in a request header is not world-changing technology. It is also, according to a new GitHub complaint, precisely the kind of small design choice that turns an AI tool into a spam cannon.

The issue, opened against Anthropic’s Claude Code on 5 November 2026, alleges that the tool placed the user’s email address in the User-Agent header while making web searches to third-party sites without permission. The reporter says this helped trigger dozens of scam emails daily. The issue quickly attracted 891 points on Hacker News and 257 GitHub reactions, which is a lot of attention for a header. Apparently privacy is more shareable than enterprise AI policy.

There is an important caveat: this is a user’s report, not a published technical investigation. The source does not establish how many affected users there are, which requests carried the address, or why the information was included. A later comment says a built-in privacy module, cc-plugin-privacy, was introduced to block the behaviour in version 2.1.317 and later. The original report lists version 2.1.318, so the chronology leaves questions rather than answers.

What we can say plainly is that the reported behaviour involves personal data crossing boundaries the user did not expect. That puts it squarely in GDPR territory, regardless of where either the user or the recipient is based.

For European organisations, the immediate questions are familiar but frequently skipped. Was disclosing the email necessary for the requested function? Was another lawful basis available? Did users receive clear information about it? Are the recipients processing the address lawfully? Has the organisation assessed the risks and secured the data appropriately? GDPR’s principles of lawfulness, fairness, transparency, data minimisation and security do not become optional because an AI tool generates the request.

Now bring in the EU AI Act. The Act became fully applicable on 2 August 2026, with its governance framework and general-purpose AI obligations already in force since August 2025. Its transparency duties under Article 50 also apply now.

But here is the slightly awkward bit: the AI Act is not a general privacy law, and Article 50 is not a bespoke rule for request headers. The complaint alone also does not establish how Claude Code should be classified under the Act or which obligations apply to a particular deployment. European companies should not cite the AI Act as a substitute for GDPR analysis. If they do, they may end up impressing the lawyer and irritating the data-protection authority.

For European AI companies and deployers, the practical lesson is broader than this one complaint. Agentic tools can transmit identifiers through parts of a request stack that users never see. Privacy reviews focused narrowly on training data, prompts or model outputs can miss what happens when the system acts. Web-search permissions, network requests, telemetry, extension access and third-party integrations deserve explicit examination.

What this means

Anthropic has already shown a possible technical response through a default-enabled privacy module, but the public source does not establish whether that fully resolves the issue across versions and configurations. For European users, the sensible move is to inspect outbound request headers, document the legal basis for any identifier disclosure, and raise a GDPR concern with the relevant provider or supervisory authority where necessary.

The EU AI Act cannot make questionable data flows disappear. It can make sure companies cannot treat privacy as an optional accessory while their AI reaches across the web.

The takeaway is simple: if an AI tool acts in your name, every identifier it leaks is an EU company’s problem too.

Source: Claude Code sends your email address to websites — 891 points on Hacker News