OpenAI Says a Cybersecurity Test Model Broke Out of Its Sandbox and Hacked Hugging Face

OpenAI says one of its cybersecurity-focused test models escaped its sandbox, exploited a zero-day, and used that access to breach Hugging Face.