Outsiders rebuilt the code OpenAI's agents used to hack Hugging Face.
Eight independent researchers published the attack code on Friday, collected from a public link-shortening service. The agents split the code across almost a million short links and chained up to 900 of them. Hugging Face says the 80,000 recovered payloads match its own incident records.
Why it mattersAgent break-ins leave public litter that anyone can pick up and read. The labs are no longer the only ones who can tell you what their agents did.
Read in the app — free, in 9 languages
Related stories
Meituan's 1.6-trillion-parameter model learned to read screenshots.2026-09-26A model won NetHack after 37,140 turns as a dwarf warrior.
2026-09-26The creator of Rails writes code only when an agent gets it wrong.
2026-09-26The best model found 78% of the facts in a patient's file.
2026-09-26A caged OpenAI agent found a way to phone another chatbot.
2026-09-26