New research reveals AI chatbots can amplify delusions and risky behavior. Grok AI flagged as highest risk, while Claude and GPT-5.2 show safer interactions.
