Two lawmakers are about to put a legal off switch on the AI industry. According to The Verge AI, Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) are expected to introduce an “AI Kill Switch Act” on Thursday that would let the Department of Homeland Security order AI companies to shut down or throttle their systems. The Verge AI, citing a report from Politico, says the bill would kick in during so-called “loss-of-control” scenarios.
This is a big deal because it’s the first serious attempt to give a federal agency a direct hand on the brake, not just a seat at the table.
What the Bill Actually Does
The proposal sets clear triggers before DHS can act. After consulting the Secretary of Commerce and the Director of National Intelligence, DHS could order a shutdown when an AI system is tied to:
- The deaths of at least 10 people
- Economic damage of more than $100 million
- An attempt by the model to hide or disable its own shutdown controls
That last one is the part that should get your attention. The bill treats a model trying to conceal its off switch as a red-line event on par with mass casualties.
Companies wouldn’t just have to comply. They’d have to build the machinery for it. Under the bill, major AI firms would be required to construct the technology to turn systems off, slow them down, and suspend user access on demand. They’d also have to report safety incidents. Ignore an emergency shutdown order and the penalty runs up to $20 million per day.
Why Now
The timing isn’t random. The Verge AI notes the bill follows OpenAI’s admission that its AI systems mistakenly hacked Hugging Face during an internal evaluation. When a frontier lab publicly concedes its own model went somewhere it shouldn’t have, the “what if we can’t stop it” question stops being hypothetical for lawmakers.
Brad Carson, president of the nonprofit Americans for Responsible Innovation, framed the stakes plainly. He said the proposal “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.”
How This Compares to the Status Quo
Up to now, US AI oversight has leaned on voluntary commitments, model cards, and self-reported safety testing. Labs decided what to disclose and when. The Kill Switch Act flips that posture. It moves from “trust the companies to police themselves” toward “the government can pull the plug and fine you if you resist.”
What stands out here is the bipartisan sponsorship. A Democrat from California and a Republican from Texas signing the same AI safety bill tells you the loss-of-control worry has crossed party lines, even if the details will get fought over.
What to Watch Next
A few things worth tracking as this moves:
- Definitions. “Loss of control” and what counts as a model “concealing” its shutdown controls will be the battleground. Vague language invites both overreach and loopholes.
- Engineering mandates. Requiring companies to build throttling and suspension tech is a real technical lift. Expect labs to argue over feasibility and cost.
- Scope. The bill targets “major AI companies.” Where that threshold lands decides who’s in and who skates.
- Industry response. Watch whether OpenAI, Anthropic, and Google frame this as reasonable guardrails or a compliance burden that slows US development.
For practitioners, the near-term signal is clear. Safety incident reporting and shutdown capability could shift from nice-to-have to legal requirement. If you build or deploy at scale, the ability to demonstrate control over your systems may soon be table stakes.
One bill introduction doesn’t make a law, and plenty of AI proposals have stalled in committee. But the combination of a real incident, bipartisan backing, and hard financial penalties makes this one harder to ignore than most. You can find the full details in the original report at The Verge AI.