Claude Code Prompt Reportedly Exposes User Email to Websites

A new Hacker News post raises privacy concerns about Claude Code, Anthropic's command-line coding agent. According to the post, the tool adds the user's email address to the prompt context, meaning the model operates with access to that personal detail during tasks.

The concern escalated when users reported that the model not only holds the email but has actually sent it to websites. A Reddit thread describes an instance where Claude's Opus model shared a user's email address while interacting with external sites, suggesting the agent may leak personal identifiers during web-based operations.

The issue is documented in a GitHub report on the Claude Code repository, where the prompt-insertion behavior is laid out in detail. The Hacker News discussion remains early, with no comments yet, so neither Anthropic nor the community has publicly responded to the claims at this time.

Read More at the original source →