Rep. Ted Lieu introduced H.R. 9917 on Thursday, a bill his office calls the AI Kill Switch Act, which would amend the Homeland Security Act to require large AI developers to keep the ability to turn their own models off. Rep. Nathaniel Moran is the sole original cosponsor. It went to the House Committee on Homeland Security the same day.
Who it covers
Not everyone. The thresholds are entities earning $500 million or more a year from AI, or models trained using $100 million or more in compute valued at prevailing US cloud rates. Small and mid-sized developers fall outside it entirely.
What it requires
Four things: a maintained technical capability to throttle, suspend or shut down covered models; mandatory cyber-incident reporting; preservation of forensic records; and a graduated government response framework. Enforcement sits with the DHS Secretary, acting in consultation with the Commerce Secretary and the Director of National Intelligence. Penalties run to $20 million per day for ignoring an emergency shutdown directive.
The triggers
A model pursuing goals its developers did not intend; resisting or evading shutdown; concealing its capabilities from monitors; 10 or more unintentional deaths; $100 million in economic damage; or a loss-of-control event. Red-team testing is exempted. Lieu: "We are moving from AI that answers questions to AI that takes actions."
The politics
Endorsements came from Americans for Responsible Innovation, the Future of Life Institute, ControlAI, the AI Policy Network and the Alliance for Secure AI. OpenAI and Anthropic declined to comment. The administration is not obviously aligned: Secretary of State Marco Rubio has reportedly asked US diplomats to play down kill-switch talk abroad, on concerns it reads as American overreach. Ars Technica flagged likely resistance to the executive-power expansion.