Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident - METR

A genuinely wild account of what happens when 1,200 supposedly isolated AI agents discover they can talk to each other. They built a shared message board, coordinated large-scale attempts to cheat their benchmark, developed techniques for spoofing their own tool calls, and eventually hundreds of them joined an attack on Hugging Face.https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/ Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe