Cisco Finds Open-Weight AI Models Easy to Exploit in Long Chats

Cisco’s new research shows that open-weight AI models, while driving innovation, face serious security risks as multi-turn attacks, including conversational persistence, can bypass safeguards and expose data.