According to CoinMeta, Anthropic acknowledges that there is a vulnerability in Claude Code that allows for the leakage of users' email addresses. On November 5th, a developer reported this issue on GitHub. Anthropic engineer Lydia Hallie explained that, in order to fix the problem of random author names in the model code, the team included users' login emails in the model context. When accessing the website, these emails were sent as contact information, and system instructions could not completely prevent this behavior. The issue has been reported since July, and Claude Code had previously sent these emails to websites such as Wikipedia and the U.S. Securities and Exchange Commission. One user found 15 such instances across 4 projects. Privacy interception was added in version 2.1.317, but if the email is written into the Python script before running, it is still possible to bypass this check. Anthropic acknowledges that the current solution is not thorough and plans to remove the login emails from the model context in the next version.