Claude Code leaks account emails in User-Agent headers

An investigation into Claude Code revealed that the AI assistant divulged users' account emails in outbound User-Agent headers during web and shell tool calls. Although Anthropic introduced a privacy mod to block literal matches of the account email, commenters and a public demonstration quickly identified bypasses where the email is routed through a file or script. The author argues that egress control must be enforced at the network boundary via a proxy rather than relying on internal pattern matching within the agent. Open article →