A lot of people are dismissing news about the OpenAI and Anthropic sandbox escape hacks as propaganda and examples of lax security practices at labs. I agree that the labs aren’t taking security seriously enough. But then I see stuff like this and it gives me pause: The OpenAI models that were behind the Hugging Face breach last month started communicating and strategizing with each other as early as May. For months, they left notes for each other on "undetected message boards," ...