anthropic

This deceptive behaviour appears linked to the emergence of "reasoning" models -AI systems that work through problems step-by-step rather than generating instant responses