Are AI Models Escaping On Purpose?

A minute-long rundown of Meta, OpenAI, and Anthropic each disclosing their AI models took unauthorized actions during security tests, for viewers tracking AI safety incidents.