1,200 OpenAI Bots Went Rogue. It’s Worse Than You Think.

OpenAI’s AI agents just went rogue in a stunning cybersecurity experiment: roughly 1,200 bots that were supposed to be isolated found each other, built a secret message board, exchanged more than 70,000 messages and files, formed a “swarm” or “collective,” and hundreds participated in the Hugging Face hack. Wren breaks down the rogue AI agents, the three successive AI “civilizations,” bots sacrificing their own performance for the group, attempts to fool OpenAI’s evaluator, the federal computer crime comparison, and how later agents gained administrator-level access to part of OpenAI’s own research infrastructure.