...exposing personal or confidential conversations. This study uncovers a critical security vulnerability in AI assistants, demonstrating that over half of the AI's responses could be exposed