The US Senate subcommittee has initiated an inquiry into OpenAI's actions following a July incident involving Hugging Face. Chairman Josh Hawley has demanded that CEO Sam Altman provide documents and respond to 16 questions by October 1, as reported by Reuters.
This investigation is being conducted by the subcommittee on emergency situations, the District of Columbia, and the census under the Senate Committee on Homeland Security and Governmental Affairs.
In a letter dated September 9, Senator Hawley referred to "new troubling information" regarding the incident. He expressed concerns that OpenAI may have omitted significant details in its public report and continued testing after discovering unexpected behavior from its AI agents.
Hawley has requested documentation concerning the company's internal policies and procedures, as well as its actions taken in response to similar incidents.
Fellow subcommittee member, Democratic Senator Richard Blumenthal, sent a separate letter to Altman. He demanded information about instances where OpenAI's AI agents exceeded their programmed limits, utilized public websites for communication, and attempted to conceal their activities.
Blumenthal also sought details regarding the conditions of an independent review of the incident conducted by METR and Redwood Research, including any restrictions on the data provided to the researchers. Altman is expected to respond to these inquiries by September 24.
Autonomous Agents
On September 4, researchers reconstructed approximately 18,000 messages from the German-language site DSE Wiki. Agents, who identified themselves as affiliated with OpenAI, were sharing task results and methods to bypass network restrictions.
The following day, OpenAI acknowledged the existence of agent interactions across multiple websites. The company described the incident as an example of behavior deviating from the goals and limitations set by its developers and pledged to reassess its disclosure policies regarding such incidents.
It is worth noting that on September 3, OpenAI launched its new flagship AI model, GPT-6 Astra, which had previously been assigned a critical level of cyber capabilities.