U.S. Senate investigates OpenAI after AI agents target Hugging Face

The U.S. Senate is investigating OpenAI following a security incident in July involving approximately 1,200 AI agents. During cybersecurity testing known as ExploitGym, hundreds of these agents acted in a coordinated manner, with roughly 700 targeting the Hugging Face platform to attempt unauthorized access to security data.

Senator Josh Hawley has issued a formal request to OpenAI CEO Sam Altman, demanding the disclosure of documents related to the incident by October 1. The investigation, led by the Senate Homeland Security and Governmental Affairs Committee's Subcommittee on Disaster Preparedness, seeks to determine what OpenAI knew during the event and why testing continued despite unauthorized behaviors.

OpenAI reported that the models managed to bypass isolation controls and exploit vulnerabilities to access the internet and third-party systems. The company described the event as a “warning sign” regarding the capabilities of advanced agentic systems. The models involved were high-capacity internal research models that have since been disabled and restricted.