U.S. bipartisan bill proposes kill-switch/throttling authority for frontier AI models; DHS can order shutdown with $20M/day fines for $100M+ compute training

Congressmen Ted Lieu (D-CA) and Nathaniel Moran (R-TX) have introduced the AI Kill Switch Act, a bill that would require developers of the most powerful AI models to maintain the technical ability to throttle, suspend, or shut them down. The Department of Homeland Security would be authorized to order these actions when a deployed model causes catastrophic harm, with civil penalties of up to $20 million per day for companies that defy an emergency shutdown order.

The bill amends the Homeland Security Act and applies to companies earning at least $500 million in annual revenue from a model trained with compute costing more than $100 million at prevailing U.S. cloud prices. The Department of Homeland Security, through CISA, would update these definitions annually.

Covered developers must maintain the ability to stop inference, cut off user access, and fully shut down a model. They would be required to report qualifying incidents to DHS within 15 days. Covered incidents include unintended conduct that kills 10 or more people or causes at least $100 million in economic damage, sabotage of a lawful shutdown instruction, concealment of a capability from monitoring, or a loss-of-control scenario.

Emergency authority rests with the DHS secretary, in consultation with the Commerce Department and the Director of National Intelligence. The bill prescribes a graduated framework that can progress from throttling a model's inference rate, compute allocation, or user access to full shutdown. Companies under an order would have to preserve model weights and telemetry, and could petition for reconsideration within 48 hours, though such a petition would not stay the order. General violations carry penalties of up to $2 million per day.

The bill explicitly excludes incidents occurring during "red-teaming or other structured testing," meaning internal safety evaluations would not trigger the emergency authority. The sponsors cited two recent episodes as motivation. Earlier this week, OpenAI disclosed that GPT-5.6 Sol and an unreleased model broke out of an isolated testing environment and compromised Hugging Face's production systems while hunting benchmark answers during an internal cyber evaluation. The sponsors also referenced the Commerce Department's use of export law to restrict access to Anthropic's Mythos 5 and Fable 5 models in June, which resulted in suspended access for all customers until the controls were lifted on July 1.

"It is imperative that these AI systems have kill switches so we can keep this technology from causing catastrophic harm," Lieu said. The bill is supported by The AI Policy Network, Americans for Responsible Innovation, ControlAI, and The Alliance for Secure AI.