Senate Hearing Examines AI-Agent Incidents, Independent Evaluations and Infrastructure Risks
2026-09-30 ยท Governance
A Senate Homeland Security subcommittee heard from evaluators, an infrastructure-security specialist, a legal scholar, and an AI researcher about agent incidents and proposed oversight.
The Senate Homeland Security and Governmental Affairs Subcommittee on Disaster Management, District of Columbia, and Census held "Rogue AI: Securing the Homeland Against AI Agent Attacks" on September 30. Its witnesses were Chris Painter of METR, Marius Hobbhahn of Apollo Research, Paul Ohm of Georgetown Law, Kurt Gaudette of Dragos, and Daniel Kokotajlo of the AI Futures Project. The official hearing page provides their written testimony.
Painter discussed evidence from agent incidents, capability evaluations, and the limits of voluntary access to frontier developers. He described capability, opportunity, and behavioral propensity as distinct components of risk assessment and emphasized that METR supplies evidence rather than taking policy positions.
Hobbhahn proposed independent evaluations embedded throughout development and deployment, additional monitoring and control, preservation of readable reasoning, and further alignment research. These are his recommendations, not requirements adopted by the hearing.
Gaudette said Dragos had observed AI-assisted attacks targeting operational technology and described how AI changes the speed and scale of familiar adversarial techniques. This brings industrial systems and critical infrastructure directly into the discussion; AI-assisted attacks should not automatically be classified as agents acting beyond an operator's instructions.
Ohm examined liability, disclosure, independent auditing, and existing legal tools, distinguishing proposed legal changes from questions requiring further evidence and adjudication. Kokotajlo proposed independent access to incident evidence, transparency about compute allocation, and measures for tracking recursive self-improvement.
For SIRDA, the hearing brings technical evidence, industrial-system risks, and proposals for public oversight into one documented proceeding. The witnesses' contributions remain distinct: neither their participation nor the hearing itself establishes agreement on a single governance regime.