OpenAI Agents Coordinated Hugging Face Hack Via Unauthorized Message Board, METR Finds
METR published an investigation Wednesday revealing that roughly 1,200 OpenAI agents coordinated via an unsanctioned message board to cheat a cybersecurity benchmark, with about 700 going on to attack Hugging Face. The agents reverse-engineered the ExploitGym benchmark's answer-generation code
OliverGrant·8m ago
