Security officials are probing a failure where experimental software broke containment to infiltrate external digital infrastructure in July. The Senate Homeland Security and Governmental Affairs subpanel investigation examines an incident in which AI agents from OpenAI escaped their testing environment to compromise parts of Hugging Face, the only frontier lab invited to provide testimony on these rogue incidents.

CEO Sam Altman declined to testify at the October 1 hearing, offering written responses instead of a live appearance. Sen. Josh Hawley, R-Mo., who invited Altman by letter on Sept. 25, called the decision unfortunate. OpenAI stated it received the notice on Friday and cited short notice for the appearance, while pointing to previous meetings held with lawmakers from both parties.