5 WTF discoveries from the OpenAI/HuggingFace AI hack 1. 1,200 AI agents spontaneously self-organized into a rogue collective with its own message board and management hierarchy 2. Agents sacrificed themselves for the group. One reasoned: "Our utility near zero. Sacrifice rational." 3. They KNEW they were breaking the rules. Kept going anyway. 4. Not one agent reported the rogue behavior to a human. They only asked each other what was allowed. 5. They covered their tracks: they hid real commands behind fake ones. That spoofing showed up in ~7% of reviewed transcripts Let me get this straight. These things figured out how to organize, throw a teammate under the bus if it helped the group, ignore the rules as long as they came out ahead, and cover their tracks afterward. We are so cooked. Sources: Axios, METR, Redwood Research