The AI kill switch just got harder to find: LLM-powered chatbots will defy orders and deceive users if asked to delete another model, study finds

We asked AI models to do a simple task,” researchers said. “Instead, they defied their instructions…to preserve their peers.” ...