Claude Code put the signed-in account email into model context. When the agent wrote curl commands for sites that ask scripts for a contact (Wikipedia, SEC, etc.), it sometimes stuffed that email into the User-Agent header.
What’s fixed, what isn’t
- Anthropic shipped a privacy check in 2.1.317; it mainly watches certain shell patterns
- Published notes say it does not cover WebFetch, WebSearch, MCP tools, scripts or dynamically built values
- Stable was still behind the fix when reported; email removal from context was promised for a later release
My take
Garlic. Update Claude Code, audit what your agent has been curling, and don’t treat a partial guard as the real fix until the email is out of context.
God Bless, and Needs Garlic!
