Image: GitHub
Why it mattersA security or privacy team running Claude Code on a shared account should check outgoing User-Agent strings today. The report is unconfirmed by Anthropic.
A coding agent that makes web requests on a user's behalf is a privacy attack surface. On 5 November 2026 a user filed GitHub issue 107757 against anthropics/claude-code. He says Claude Code sent his email address to third-party sites in the User-Agent header. The requests were web searches, done without his permission.
The report was discussed on Hacker News on the same day. The HN API reports 159 points and 38 comments. The title is "Claude Code sends your email address to websites".
What the issue says
The reporter writes that Claude Code "divulged my email to arbitrary 3rd parties in the User-Agent request header". He says it happened during web searches, "without my permission". He also says the fallout is "dozens of scam emails per day" at the leaked address.
The environment on the bug report is Claude Code version 2.1.318 on macOS, in a tmux session. The reporter does not name the sites that received the email. He does not post a captured request. He calls for the behaviour to be removed from the tool entirely rather than hidden behind a setting.
As of 5 November 2026, Anthropic has not replied on the issue. No fix branch or pull request is linked. The issue state is Open with no assignee.
What a team should do today
The report is a single user's claim. The claim is not verified by an independent capture, and Anthropic has not confirmed or denied it. A team cannot treat the specific behaviour as proven. A team can still check it.
Three things to do. First, send a Claude Code web search to a server a team controls and log the User-Agent header. That takes one netcat listener or a one-line HTTP server. If the reporter is right, the email sits in the header. Second, check whether the leaked address, if any, is a personal address or a shared one. A personal address behind a corporate Claude Code account is a different problem to a shared mailbox. Third, decide whether to route Claude Code's HTTP egress through a proxy that strips the User-Agent. That is not a fix. It is a stop-gap until Anthropic responds.
A broader lesson sits under this one. A coding agent is a client program that makes outbound requests on the user's behalf. A client program's User-Agent header reaches every endpoint it talks to, which includes any server a search returns. A rule for the whole class: audit what identifying strings the agent attaches to outbound requests before adopting a new version.
The claim is from one user. The issue stays Open until Anthropic acknowledges or denies. This item updates only when either is published.
Source
GitHub issue, "Bug Report: Privacy violation in User-Agent header". Hacker News discussion.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
