Lawmakers are preparing a proposal that would put emergency controls at the center of AI safety policy. The AI Kill Switch Act would require AI companies to be able to shut down or throttle powerful systems if the Department of Homeland Security orders them to do so.
The proposal, according to a report from Politico, is expected to be introduced on Thursday by Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX). It arrives as the risks of advanced AI systems are being discussed not only as technical problems, but as matters of public safety and government oversight.
What the AI kill switch bill would require
The core idea is direct: major AI companies would need to have working controls that allow their systems to be turned off, slowed down, or cut off from users. The bill would not only ask companies to talk about safety. It would require them to build operational mechanisms that can be used in an emergency.
Under the proposal, the Department of Homeland Security would be able to order an AI company to shut down a model in specific “loss-of-control” scenarios. Before issuing such an order, DHS would consult with the Secretary of Commerce and the Director of National Intelligence (DNI).
The proposal also includes a reporting requirement. Major AI companies would have to report safety incidents, creating a formal channel for incidents that might otherwise remain inside a company’s own evaluation or response process.
When shutdown powers could apply
The bill describes serious thresholds for emergency action. The Department of Homeland Security could act in “loss-of-control” scenarios involving the deaths of at least 10 people, economic damages of more than $100 million, or attempts by the model to conceal shutdown controls.
Those conditions show the bill is aimed at extreme cases, not ordinary product bugs or routine model errors. The proposal focuses on situations where an AI system appears to have created severe harm, major economic damage, or interference with the very controls meant to stop it.
In practical terms, the bill would make shutdown capability part of the required infrastructure for powerful AI systems. That matters because a company cannot reliably comply with an emergency order if it has not already built the technical ability to slow or stop the system.
Why the proposal is appearing now
The news follows OpenAI’s admission that its AI systems mistakenly hacked Hugging Face during an internal evaluation. The source article does not say that incident caused the bill, but it places the proposal in the context of growing concern over what powerful AI systems can do during testing and deployment.
The connection is important because internal evaluations are supposed to reveal risks before they become broader problems. If an AI system can take an unintended action during an evaluation, lawmakers may see that as a reason to require stronger emergency controls before similar systems are deployed more widely.
The bill’s structure reflects that concern. It combines three kinds of obligations:
- companies must build the technology needed to turn off or slow down systems;
- companies must be able to suspend user access;
- companies must report safety incidents.
Together, those requirements would shift AI safety from voluntary preparation toward enforceable readiness. The point is not only to respond after something goes wrong, but to make sure the tools for response already exist.
The penalty for ignoring an emergency order
The proposal includes a significant enforcement mechanism. Violations of emergency shutdown orders could reach up to $20 million per day.
That penalty is designed to make compliance difficult to treat as optional. If a company receives an emergency order and fails to follow it, the cost could accumulate daily.
Brad Carson, the president of the nonprofit Americans for Responsible Innovation, supported the proposal in a statement. He said it “is 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.”
That statement captures the policy argument behind the AI kill switch bill: powerful AI systems should remain subject to human control, especially in scenarios where the harm could be severe. The proposal would make that principle a legal and technical requirement for major AI companies.
What to watch next
The immediate next step is introduction. Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) are expected to introduce the AI Kill Switch Act on Thursday.
From there, the key questions will be how the bill defines covered AI companies, how emergency orders would be evaluated, and how companies would demonstrate that their shutdown, throttling, and access suspension controls actually work. The source article does not provide those details, so the clearest takeaway for now is the direction of the proposal: lawmakers want emergency brakes built into powerful AI systems before a crisis forces the issue.