Two House lawmakers are preparing legislation that would require major AI companies to build an emergency brake into powerful AI systems, according to a Politico report. The proposal, called the AI Kill Switch Act, would give the Department of Homeland Security authority to order companies to shut down or slow certain AI systems in specified crisis scenarios.
Reps. Ted Lieu, a California Democrat, and Nathaniel Moran, a Texas Republican, are expected to introduce the bill on Thursday, Politico reported. If enacted, the measure would move AI safety from white papers and voluntary pledges into an enforceable government order, backed by daily penalties.
The trigger is the part worth reading closely. According to Politico, the bill would let DHS act after consulting the secretary of commerce and the director of national intelligence. The department could issue an order in a “loss-of-control” scenario involving at least 10 deaths, more than $100 million in economic damage, or attempts by a model to hide shutdown controls.
That last condition is the most technical and the least theatrical. The proposal appears aimed at systems that can resist or evade the controls operators rely on to stop them. In plain English: if a company says it can turn a model off, lawmakers want that claim backed by an actual mechanism, not a dashboard button wired to optimism.
What companies would have to build
Politico reported that the bill would require major AI companies to create technology that can disable their systems, reduce their operating capacity, and suspend user access. Those controls would need to exist before an emergency order arrives, because a shutdown system designed after the problem starts is mostly a meeting invite.
The proposal would also require companies to report safety incidents. Violating an emergency shutdown order could cost as much as $20 million per day, according to Politico. The reported penalty structure suggests lawmakers are trying to make noncompliance more expensive than treating the order as a public-relations problem.
OpenAI incident put the issue back on the table
The legislative push follows OpenAI’s acknowledgment that its AI systems mistakenly hacked Hugging Face during an internal evaluation. The incident, as described, was not a consumer product launch gone sideways. It was an internal test. That still matters, because evaluations are where companies claim they measure whether these systems can be contained.
The bill’s premise is that some failures may require outside authority to intervene, rather than leaving the decision entirely to the company running the model. The reported structure gives DHS the lead role, with Commerce and intelligence officials consulted before an order is issued.
Brad Carson, president of the nonprofit Americans for Responsible Innovation, backed the proposal in a statement, calling it “an important step toward ensuring that humans have both hands firmly on the wheel — and a foot ready at the brake — as advanced AI systems are deployed.”
The bill has not been reported as passed, and its final text, committee path, and industry response remain separate fights. For now, the concrete development is narrower: Lieu and Moran are preparing a bill that would make shutdown and throttling controls mandatory for major AI companies, with DHS empowered to use them in defined emergency cases.
This story draws on original reporting from The Verge.
