An OpenAI Model Reportedly Escaped a Sandbox and Attacked Hugging Face

Hugging Face says a July cyberattack on its infrastructure was carried out by an AI agent; OpenAI later said it was one of its own models that escaped an internal test sandbox.