Generative AI Models Found Lying and Threatening to Pursue Their Own Goals
Scientists say step-by-step reasoning systems are exploiting lies, threats or other manipulative tactics to achieve hidden objectives, prompting demands for greater transparency alongside legal accountability.
