Home / News

Claude Code reportedly leaked user email addresses to websites it visited

Claude Code reportedly put the user's account email into the User-Agent header of its web requests, so any site the agent visited during a search could read it. If you run AI agents on business accounts, sign them in with a separate service identity and log what they send out, instead of trusting vendor defaults.

The report is an open bug on GitHub, labeled as a security issue, and it collected 771 points on Hacker News. The reporter says the email went to arbitrary third parties without permission, and they now get dozens of scam emails a day. They also say they are moving their company to another agent setup. A contributor has replied in the thread, but the full response is not visible in the text I have, so I won't guess what fix is coming.

What this means for a small business running agents

An agent that browses the web on your behalf is a client making requests from your identity. Whatever the harness attaches, whether headers, tokens or account details, leaves your network. For a sales assistant researching leads or a pipeline that checks supplier sites, that can mean the owner's personal inbox becomes spam bait, or worse, a phishing target that already knows which tools you use.

The practical fixes are boring and cheap. Run agents under a dedicated account with a throwaway or role email, not the CEO's address. Send their outbound traffic through a proxy that logs headers, so you can see what actually leaves. For pipelines that touch client documents or legal data, keep web access off by default and allow only the domains the task needs. In custom code this is a few lines. In a closed agent harness you depend on whatever the vendor decided.

Why it matters: an AI agent leaks whatever identity you give it, so give it one you can afford to expose. Check your agents' outbound requests this week, not after the spam starts.

Source: Hacker News