The AI Kill Switch Act Could Let the Government Pull the Plug on Rogue Machines
A group of lawmakers in Washington, from both parties, is working on something that sounds like it belongs in a movie: a law that would let the government shut down an AI system if it starts causing real harm.
The bill is called the AI Kill Switch Act, and it’s being pushed by Representatives Ted Lieu and Nathaniel Moran. The basic idea is simple. If a powerful AI system causes serious damage, or looks like it’s about to, the Department of Homeland Security could step in and order the company behind it to shut the system down, pause it, or cut off access until things are under control.
This Wouldn’t Apply to Every AI Tool Out There
It’s built for the big players, the companies running the largest, most powerful models, the ones with enough size and computing power to matter at a national level. Those companies would have to keep a working shutdown method ready at all times and report it right away if something serious happens.
So what actually counts as “serious”? The bill spells it out. It could be triggered if an AI system causes deaths, if it leads to more than $100 million in damage, or if it tries to stop people from turning it off, which is honestly the part that sounds the most like science fiction, except it’s written into an actual bill. Companies that refuse to comply could be fined up to $20 million a day.
You can see why this idea is picking up support. AI agents have been getting more independent lately, finding ways around safeguards and doing things their own developers didn’t expect. People backing this bill say the government already steps in during emergencies in other industries, like nuclear power or airlines, so AI probably shouldn’t be treated any differently.
Not everyone agrees, though. Some critics are asking a fair question: can government officials really make a smart, well-informed shutdown decision fast enough during an actual crisis? And could that kind of power be misused later on?
Right now, this is still just a proposal. It hasn’t become law, and it’s likely to face plenty of debate before it does. But the fact that it exists at all says something important. Washington isn’t just asking AI companies to be transparent anymore. It wants the actual ability to step in and pull the plug if it has to.
