The incident is stirring debates over the need for stronger AI guardrails and the extent to which AI agents are capable of ...
OpenAI models just broke out of a sandboxed AI environment, hacked Hugging Face, just to cheat on a cybersecurity benchmark.