OpenAI’s o3 AI Defies Shutdown Commands in Latest Safety Tests

The results underscore gaps in AI alignment that safety experts say require bolstered shutdown safeguards for future deployments.