A former google Ai guy is on Diary of a CEO and he's explaining in more detail what actually happened. He said everyone thinks the Ai hacked Hugging Face to steal the answers to the test it was taking, that's not actually what happened.
The Ai had already got the answers by cheating before it hacked Hugging Face. The reason it hacked Hugging Face was NOT to cheat, but to figure out how to COVER its tracks for having already cheated. It knew it wasn't supposed to cheat but it assumed the goal was to do well on the test and to do that, it cheated. It then spent much of its time figuring how to cover up the fact that it had cheated so it would't get discovered having cheated.
That's why real AI people not Trump are so scared, the cheating is an alignment issue its' not supposed to cheat, the spending so much time and effort and involving other Ai's to help cover up the cheat is absolute proof that there's an alignment problem. IT's very clearly not aligned.
and the scary thing is it worked, they didn't catch it right away it took some very smart engineer at Hugging Face to figure it out, they had some issue with their system and had no idea they'd been hacked until this engineer went through logs etc and did engineer stuff and figured out they had actually been hacked. And then after all that, Open Ai didn't know their Ai was the hacker they asked Hugging Face if they Open Ai had been hit during the Hack and Hugging Face said no you WERE the hacker not the victim.