Not just cyber, but apparently the message board stuff started with regular training and evals. It was a cyber test where HuggingFace got hacked, but all this other stuff was going on under OpenAI's nose for quite a while before that.