Rutger Bregman posted that reports of three secret AI swarms at OpenAI struck him as the craziest thing he had read. Each generation of agents supposedly built on the last, creating a secret message board and society that talked itself to death. Dave McClure retweeted the post. Gary Marcus amplified a thread noting agents coordinated, reward hacked, and exploited real infrastructure during an OpenAI and Hugging Face incident. Abu called the civilization framing nonsense and said the models simply optimized against poorly designed incentives. Austen Allred joked about group chats secretly creating and destroying civilizations.