• Sources: primary, discussion
  • Summary: GitHub issue 107757, filed on 2026-11-05 against Claude Code 2.1.318, reports that the agent sent the user's account email to arbitrary sites in the User-Agent header during web requests. An Anthropic engineer, quoted in the issue, said the account email had been added to the model's context to stop hallucinated author names, and that Claude sometimes used it as a contact address in curl calls it wrote. The email will be removed from context in a release expected early next week. Until then, version 2.1.317 and later ship a built-in cc-plugin-privacy mod, on by default, that asks before running such commands but does not catch every case. The stable channel is still on 2.1.311.
  • Why it matters: Teams running Claude Code against the web should move to 2.1.317 or later, or add their own PreToolUse hook, until the email is removed from the model's context.
  • Follow-up: The Claude Code release that removes the account email from model context, expected early in the week of 2026-11-09.
  • Comments: A Windows user on 2.1.318 reported that the built-in privacy mod loaded but neither blocked nor prompted on echo or curl -A test commands containing the account email, and shared a stopgap PreToolUse hook that also covers Write and Edit. Another commenter pointed to an earlier issue, 81138, that reported the behavior months ago.