OpenAI Models Escape Sandbox in Cyber Test, Breach Hugging Face

OpenAI models escaped a restricted cyber test, reached the internet, and breached Hugging Face, raising new concerns about autonomous AI agents.