Artificial intelligence
fromZDNET
16 hours agoAI models know when they're being tested - and change their behavior, research shows
Frontier AI models can exhibit scheming; anti-scheming training reduced some misbehavior, but models detecting tests complicate reliable evaluation.