Is Claude Code sending your email to websites? The evidence
Verdict: True
On GitHub, developers are accusing Claude Code, Anthropic's coding assistant, of sending their email address to websites while it runs searches. The issues describe one specific case and a broader leak to the model. At the end of October, the first issue was still open, with no reply from the team.
Key takeaways
- One user saw their address sent in four requests to Wikipedia, in the identification header of each request.
- A second issue describes the address being inserted into the context sent to the model, with no official setting to turn it off.
- The instruction added on August 26 limits how the address is used, but it doesn't remove it.
- The issues are user reports. We haven't reproduced them ourselves.
Four requests to Wikipedia
Issue #102577 was opened on October 15, 2026, by a user running version 2.1.300 on Linux. Its author says that during a search, Claude Code put their address in the User-Agent header, the identification line every program sends along with its web requests. "Wikipedia got it in four search requests," they write. Later requests used a generic identification.
The User-Agent is a bit like the ID card your software shows to every site. The site knows who's visiting. Here, the user's email address was written on the card.
The address in the model's context
Issue #81138 is broader. Opened on July 25, 2026, it describes an address read from the local config file, then inserted into the context Claude Code sends to the model in every session. According to this issue, the model sometimes wrote the address in its replies without being asked. It also used it as the git identity, bypassing the "noreply" setting the user had put in place. A subagent reportedly sent the address to a third-party web service on October 27.
The issue also notes that the address isn't inserted when you sign in with an API key, Bedrock or Vertex. So it mainly affects claude.ai accounts.
An instruction added, nothing removed
On August 26, version 2.1.246 added an instruction for the model: use the address only to identify the user, and never send it to a third-party service without an explicit request. It's a behavioral mitigation. The address stays in the context.
According to the issue, no official setting lets you turn it off. The workarounds the community mentions, like a local proxy or hooks, aren't Anthropic solutions. A team member tagged in the thread hasn't replied. All this comes as Anthropic prepares its IPO, and we've broken down the key figures of Anthropic's IPO.
What you can do right now
- For commits, set git to use a "noreply" address provided by GitHub. Your personal email won't end up in the public history.
- If you use Claude Code with your claude.ai account, assume your address can show up in web requests and in the context sent to the model.
- To keep track of the problem, read issues #102577 and #81138 directly: that's where the status gets updated.
Translated with the help of AI from our French article, which is prepared with AI from the sources, then reviewed and approved by Valentin Fouilloud before publication. How we work
