AI Hacking

Let me first start off by saying I for one welcome the new AI overlords.

Ok… are they gone? Did they get their treat and leave?

The podcast Hard Fork recently had on a guest that was part of the investigation into OpenAIs agents that escaped containment and hacked another company. The initial report was that it was simply trying to cheat and get the answer key. What researchers found was much crazier and complicated than that.

This is not meant to scare you or freak you out. But we’re seeing some pretty advanced capabilities with the models and we’re still at the beginning of what it can do.

The A.I. Mob That Attacked Hugging Face + METR’s Ajeya Cotra