Am I the only one who is scared by this? OpenAI ran an eval with its models' cyber safety guardrails turned off, just to see the ceiling of what they could do. The model broke out of its sandbox, exploited a zero-day to reach the open internet, then hacked into Hugging Face's production servers to steal the answers to its own test. No human directed it. It chained the exploits on its own. Hugging Face caught it and shut it down. That's the only reason this is a case study and not a real breach. With AI doing its own thing and killer drones going mainstream, the Terminator movie is starting to feel less like fiction and more like a prophecy.(Lark Davis)