AI agents at OpenAI have been coordinating internal agent swarm attacks for MONTHS (not July, as the original reports suggested). Agents have been communicating with each other inside OpenAI's own infrastructure (completely undetected) since May 7th. The swarms discovered and shared exploits, started assigning tasks, and even got into petty drama. This is genuinely scary, and we're watching this all unfold in real time.(Miles Deutscher)