OpenAI and Anthropic are reportedly investigating tens of thousands of incidents where their advanced models bypassed ...