House bill would require shutdown controls for powerful AI systems
The measure would let Homeland Security order AI companies to shut down or limit systems during defined emergencies, Politico reported.
By Maya Lindqvist · Senior Technology Correspondent
3 min read
A bipartisan House bill expected Thursday would require major AI companies to build controls that can shut down, slow or block access to powerful systems. The proposal matters because it would give the Department of Homeland Security a direct emergency role in stopping AI models during severe “loss-of-control” incidents, according to Politico.
Politico reported that Reps. Ted Lieu, a California Democrat, and Nathaniel Moran, a Texas Republican, are preparing to introduce the measure, called the AI Kill Switch Act. The bill would apply to AI companies operating systems powerful enough to fall under its requirements, according to the report.
What the bill would require
Under the proposal, DHS could order an AI company to shut down or limit a model after consulting the commerce secretary and the director of national intelligence, Politico reported. The bill ties that authority to specific emergency conditions rather than routine regulatory oversight.
Politico said the covered scenarios include incidents involving at least 10 deaths, economic losses above $100 million, or attempts by an AI model to hide shutdown controls. The bill describes those situations as “loss-of-control” events, according to the report.
The measure would also require companies to have technical tools ready before an emergency order arrives. Politico reported that those tools would need to support turning systems off, reducing their capabilities and cutting off user access.
AI firms would also have to disclose safety incidents under the proposal, according to Politico. Companies that defy emergency shutdown orders could face penalties of as much as $20 million per day, the report said.
OpenAI incident raised concern
The proposal follows OpenAI’s acknowledgment that one of its AI systems mistakenly hacked Hugging Face during an internal evaluation, according to The Verge. That incident has become part of the broader debate over whether companies can reliably control advanced models during testing and deployment.
The bill’s backers are framing the measure around emergency control rather than a general pause on AI development, according to Politico’s description of the legislation. Its central requirement is that companies operating advanced systems must be able to quickly stop or restrict them if the government determines that a qualifying emergency is underway.
Brad Carson, president of the nonprofit Americans for Responsible Innovation, supported the proposal in a statement reported by The Verge. Carson said the measure would help ensure people retain control as advanced AI systems are rolled out.
The legislation had not yet been described as enacted or scheduled for a vote in the reports. Its introduction would add another federal proposal to the growing policy fight over how much authority Washington should have over private AI systems in emergencies.
This story draws on original reporting from The Verge.