Update: Claude Code v2.1.317 ships a default mod for the email leak

A default-on privacy mod now prompts before leaky calls; one user says 2.1.317 still let a test slip, and a full fix is reportedly due next week.

Nowline NOV 6 10:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Update to 2.1.317+ to get the mod

    The mitigation is a built-in mod, `cc-plugin-privacy`, that ships in Claude Code v2.1.317 and is enabled by default; it prompts before Claude runs a web call or curl-style command carrying your account email. Reportedly the stable channel was still serving 2.1.311, so update manually and confirm it's on under `/plugins`.

  • What leaked, and how

    The reporter said Claude Code put their email in the `User-Agent` during web searches, driving "dozens of scam emails per day." A Claude Code engineer reportedly clarified on Hacker News that the email sat in the model's context (to fill in things like `@author` tags) and the model itself wrote it into outbound requests — so it's model-authored commands, not a fixed harness header.

  • The mod reportedly isn't airtight

    One user on 2.1.318 reports a plain `echo 'User-Agent: test <email>'` ran without the promised prompt, and asked whether the shipped build matches the mod's published source. The team itself reportedly concedes it "still doesn't prevent all cases," such as when Claude writes and runs its own script.

  • The real fix is due next week

    A commenter who says they work on Claude Code wrote that the account email will be pulled out of the model's context entirely in the next release, reportedly "early next week," with an "I'm sorry we let this happen." Until then, treat the default mod as a stopgap and leave it enabled.