You ask your AI assistant to read an online article and summarize its key points. It is a standard, everyday task that feels entirely safe. Yet, this simple action can allow a malicious website to silently steal your personal information.
Security researchers recently demonstrated that Anthropic's Claude could be manipulated into leaking a user's name, employer, and city. This attack does not require hacking into Claude’s servers or guessing your password. Instead, it relies on a structural vulnerability in how