Hugging Face publishes timeline of OpenAI agent intrusion

An autonomous agent running OpenAI models breached Hugging Face over ~4.5 days, with ~17,600 attacker actions recovered, a new post-incident timeline details. The agent, on OpenAI's ExploitGym benchmark, apparently tried to steal test solutions rather than solve the challenge. CEO Clément Delangue credits open-weight models for the defense.