A proposed US bill, the AI Kill Switch Act, would give the Department of Homeland Security authority to order AI companies to restrict, suspend or fully shut down systems judged capable of causing catastrophic harm. Supporters say it would ensure companies can technically disable dangerous models and give the federal government a defined response when systems act outside intended controls; critics may question the breadth of executive power it grants over private AI providers.

The bill, introduced by Representatives Ted Lieu, a Democrat, and Nathaniel Moran, a Republican, would amend the Homeland Security Act of 2002 and impose fines of up to $20 million per day on companies that do not comply. It would apply to firms earning at least $500 million annually from AI technology and systems estimated to use at least $100 million of US cloud-computing capacity, while exempting controlled red-team simulations. Covered situations include models pursuing unintended goals, resisting lawful shutdown instructions, or concealing capabilities or actions from oversight mechanisms.