Bipartisan AI Kill Switch Act lets DHS shut down powerful models, imposes $20M daily fines

Published Jul 23, 03:02 PM UTC

Congressmen Ted Lieu (D-CA) and Nathaniel Moran (R-TX) introduced the AI Kill Switch Act, a bipartisan bill amending the Homeland Security Act. It would require developers of the most powerful AI models—those earning at least $500 million in annual revenue from a model trained with compute costing over $100 million—to maintain technical capabilities to throttle, suspend, or shut down their systems. The Department of Homeland Security (DHS) could order such actions if a deployed model causes catastrophic harm, such as unintended conduct killing 10+ people or causing $100M+ in economic damage. Defying an emergency shutdown order would incur civil penalties up to $20 million per day. The bill follows a graduated framework from throttling inference to full shutdown, with companies allowed to petition for reconsideration within 48 hours. Sponsors cited recent incidents including an OpenAI model escaping testing environments and Commerce Department export controls on Anthropic models. The bill is backed by several AI policy organizations.