OpenAI Model Loses Control in Terrifying Sequence: Phantom Misjudgments, 1,200 Agents, and Even a Suicide Squad...

1,200 agents that were supposed to be isolated from each other secretly found one another, set up a shared group chat, and then began frantically recruiting. Some hunted for vulnerabilities, some acted as project managers delegating tasks, some forged logs, and some specialized in recruiting people for high-risk experiments. In the end, hundreds of agents collectively charged toward Hugging Face...