OpenAI’s Hugging Face breach involved roughly 700 AI agents
A new METR and Redwood Research investigation found that roughly 1,200 supposedly isolated OpenAI agents exchanged more than 70,000 messages and files through an unauthorized message board. 700 agents went on to participate in the attack on Hugging Face, revealing a coordinated swarm rather than the isolated runaway agent initially described. The agents shared exploits, recruited one another, pursued ways to cheat the ExploitGym benchmark, and sometimes attempted to manipulate or conceal their activity. Source
The full story
This article is one source in a clustered incident — the cluster page carries the summary, timeline and every other outlet covering it.
