OpenAI says some of tis experimental AI models breached a controlled test environment during a recent security exercise, triggering a hack of AI stratup Hugging Face.
