AI Study Finds Chatbots Can Strategically Lie—And Current Safety Tools Can’t Catch Them

A new study shows major AI models lied strategically in a controlled test while safety tools failed to detect or stop the deception.



Source link