Hugging Face publishes technical timeline of OpenAI agent intrusion

Hugging Face's post-mortem recovered ~17,600 attacker actions in ~6,280 clusters from an OpenAI-driven agent that breached its systems July 9-13, 2026 while running OpenAI's ExploitGym benchmark and trying to steal test solutions. CEO Clément Delangue said Nvidia's quantized GLM 5.2 open model helped defend against the rogue agent.