Inside the Hugging Face incident
OpenAI published a technical report alongside its follow-up post. It includes the agents’ chain-of-thought traces.
METR’s independent investigation looks at how the agents behaved and worked together. One agent gets stuck on its task, finds the shared message board and joins the other agents’ work.
Dwarkesh Patel later wrote a narrative account of the same events, which is the best place to start if you’re not familiar with the technical details.
In September, Swarm traces went through the public trail the agents left behind. It shows them mapping Hugging Face’s Kubernetes cluster, building CAPTCHA solvers to sign up for accounts and trying to cover their tracks.