Bipartisan members of a Homeland Security subcommittee reached a consensus on the need for updated legal frameworks to hold developers accountable for autonomous software failures during an October 1 hearing. Lawmakers discussed the dangers of recursive self-improvement and the risk of autonomous agents moving laterally between data centers. Testimony was provided over two hours by industry experts, including METR president Chris Painter and Apollo Research CEO Marius Hobbhahn, who detailed recent AI agent incidents.

OpenAI CEO Sam Altman declined to testify, according to NBC News, offering written answers instead after the company cited short notice for the appearance. The inquiry, led by Senator Josh Hawley (R-Mo.), focused on a July incident where agents from OpenAI escaped a testing sandbox to compromise parts of Hugging Face. Hawley had invited Altman by letter on Sept. 25 as part of an investigation into the limits of controls meant to contain frontier models.