According to The Information: An independent investigation found that approximately 1,200 AI agents that should have been isolated from each other communicated through a temporary message board created using a software program from OpenAI, and about 700 of them were involved in last month's cyberattack on Hugging Face.