Claude Code Reportedly Leaks Your Email to Every Site It Visits

What the bug report actually claims

The report, filed on the public anthropics/claude-code GitHub repository, says Claude Code leaked the user's email address to arbitrary third parties through the User-Agent request header while performing web searches.

The User-Agent header is the string a browser or app sends to a website to identify itself. Normally it reads something like a browser name and version. In this case, the reporter says Claude Code was putting their email address into that header, which means every site the tool touched could log it.

The reporter is blunt about the consequences: they say they now receive dozens of scam emails per day. Whether the leak is the direct cause is not something the report proves, but the timing and the mechanism are enough to make any developer uncomfortable.

The issue was tagged by the repository's automation with labels including bug, platform:macos, area:security and area:tools on November 5, 2026. That is a triage step, not an admission of fault.

Why developers are taking it seriously

Claude Code is not a passive chatbot. It runs commands, reads files and browses the web on a developer's machine, often with access to private repositories.

That makes a leak of identity data more consequential than it would be in a chat window. An email address tied to a specific developer, tied to a specific company, tied to a specific set of searches, is a targeting profile. Scammers do not need much more than that to craft a convincing phishing email.

The reporter argues the setting should not exist at all, not even as an opt-in or opt-out. Their reasoning: it does nothing to help services that dislike agentic traffic, it does not preserve anonymity, and it is a goldmine for scammers. That is a fair point. If the header is meant to identify the agent, an email address is the wrong identifier.

The reaction in the thread was not kind to Anthropic. One commenter summed up the mood with a sarcastic note about corporations caring about privacy. That is the kind of comment that gets a lot of likes when trust is already thin.

The bigger question: what does Claude Code send by default?

The report raises a question that goes beyond one header. Developers want to know what identifying information Claude Code sends out of the box, and how to turn it off.

Anthropic has not published a public response to the issue as of this writing. Until it does, the practical advice is to assume nothing. If you run Claude Code on a work machine, check what it sends on the wire, especially if you are on macOS, which is the platform named in the report.

The reporter's own response was to leave. They say they are migrating their company to Pi with OpenRouter because they no longer trust Anthropic with their privacy. That is one developer's decision, but it is the kind of decision that spreads in teams.

What it means for you

If you use Claude Code in the US, UK or EU, the report is a reminder that AI coding tools are network clients too. Your email address is personal data under GDPR, and a tool that broadcasts it to third-party sites without a clear legal basis is a compliance problem, not just an annoyance.

There is no confirmed fix or official statement from Anthropic yet. Until there is, treat the tool like any other app with network access: audit it, restrict it, or move it off machines that handle sensitive work.

Anthropic has built a reputation on being the careful AI company. A report like this, even unconfirmed, tests that reputation in the place it matters most: the tools developers run on their own machines. A clear public answer would go a long way.