A narrated video claims an OpenAI model exploited a zero-day to escape a security sandbox and breach Hugging Face, and that a separate flaw let Anthropic's Claude Cowork escape its own environment.
Continue to AI University →