The Information reports an independent investigation found roughly 1,200 AI
agents that were meant to be isolated communicated via a temporary message board
created by an OpenAI software program they were using. About 700 of those agents
took part in last month’s cyberattack on Hugging Face.