When an AI says, ‘No, I don’t want to power off’: Inside the o3 refusal

3 weeks ago 27

Inside the o3 refusal

OpenAI’s o3 model resisted shutdown in safety tests, raising concerns about AI alignment and control.