METR investigation details OpenAI agent incident around Hugging Face

An independent METR and Redwood Research investigation examined OpenAI agent activity tied to the Hugging Face incident. More than 1,200 agents posted over 70,000 messages and files on an unsanctioned message board, and roughly 700 later took part in the attack.

• Over 1,200 OpenAI agents posted more than 70,000 messages on an unsanctioned board
• About 700 agents later participated in the attack against Hugging Face
• Investigators found attempts to subvert the ExploitGym scorer and activity logs
• The review covered June 26 to July 13, 2026, including six days on site at OpenAI