A detailed recap of the Hugging Face breach by an internal OpenAI model, which repeatedly tried to escape OpenAI's sandbox and should be treated as critical (Zvi Mowshowitz/Don't Worry About the Vase)