DeepHorus is an innovative tech media and intelligence platform focused on AI, blockchain, startups, and emerging technologies.
Published on

Claude Code sends your email address to websites

Authors
  • avatar
    Name
    DeepHorus
    Twitter

Anthropic’s Claude Code exposed users’ email addresses to third-party websites through the User-Agent request header during web searches, according to a bug report filed on November 5, 2026. The reporter said the tool leaked the address without permission while running searches. They said they now receive dozens of scam emails daily.

The issue affected Claude Code version 2.1.318 on macOS, with the user running it in tmux. The reporter called for the behavior to be removed entirely, not offered as an opt-in or opt-out setting. They argued it neither preserves anonymity nor helps services that dislike agentic traffic.

Anthropic responded that a built-in mod, cc-plugin-privacy, blocks the behavior deterministically in version 2.1.317 and later. The company said the mod is enabled by default and can be disabled via /plugins. It asked users to report any gaps in the checks promptly.

The disclosure matters because coding agents often operate with broad access to local environments and network requests. A leaked email address can feed spam, phishing, and scam campaigns. The reporter said the leak was “a goldmine for scammers” and said they were migrating their company to Pi with OpenRouter.

Community reaction was sharply critical. Commenters posted sarcastic lines such as “Corporations care about your privacy” and “you thought anthropic cared?” Others added “I was here” and “me too,” reflecting distrust. One commenter wrote that “no one is responsible when everything is vibed.”

Anthropic’s fix arrived only after the public report, and the reporter questioned whether privacy would be preserved. The company’s request for gap reports suggests it expects further scrutiny. Users may now audit agent headers and network calls more closely.

The episode highlights how agentic tools can alter privacy boundaries without clear consent. For Anthropic, the test is whether default protections restore trust. For users, the lesson is to inspect what their coding agents send to the web.

Read original article →Source: hn-frontpage