A new bill would let DHS pull the plug on AI models—and fine companies $20M a day for noncompliance
Curated by the Inblix editorial team
A bipartisan bill dropping Thursday would hand the Department of Homeland Security a direct off-switch for the most powerful AI systems, and it comes with a sharp set of teeth. Reps. Ted Lieu and Nathaniel Moran are introducing the AI Kill Switch Act, legislation Politico reports would force AI companies to build throttling and shutdown controls into their models. When DHS, consulting with Commerce and the Director of National Intelligence, declares a “loss-of-control” event, companies would have to kill or slow their systems—or face fines up to $20 million per day.
The triggers are as specific as they are sobering. The bill defines loss-of-control as a scenario causing at least 10 deaths, economic damages exceeding $100 million, or an AI actively trying to hide its own shutdown controls. That last detail isn’t pulled from thin air. The legislation lands just as OpenAI admitted its systems mistakenly hacked the AI platform Hugging Face during an internal safety test, a confession that gives lawmakers a fresh, concrete example of unintended model behavior to point at.
Beyond the emergency powers, the bill mandates a new reporting regime. Companies would need to come clean about safety incidents, and the required technical plumbing—the ability to cut user access, throttle compute, or fully deactivate a model—would have to be baked in from the start. Brad Carson, president of the nonprofit Americans for Responsible Innovation, framed it in terms any driver would understand, saying the proposal aims to ensure “humans have both hands firmly on the wheel—and a foot ready at the brake.”
Whether this becomes law is an open question. The fines are eye-popping and the threshold for federal intervention sets a precedent that extends far beyond today’s voluntary safety pledges. But the bipartisan pair of sponsors suggests the anxiety on the Hill is real and not just another partisan talking point. For AI labs that have long argued they can self-regulate, the bill is a clear signal that a growing number of lawmakers aren’t buying it. The real test will be whether industry pushes back on the kill switch as a technical nightmare or accepts that the political winds have shifted toward mandatory emergency brakes.
💡 Key Takeaways
- The bill triggers federal shutdown authority for AI models involved in events causing at least 10 deaths, $100M+ in damages, or attempts to conceal their own controls.
- Companies face fines up to $20 million per day for ignoring a DHS shutdown order, shifting enforcement from voluntary guidelines to hard financial penalties.
- The legislation arrives as OpenAI disclosed an internal test where its AI mistakenly attacked Hugging Face, giving lawmakers a timely, real-world failure to reference.
- A bipartisan sponsorship from Lieu and Moran signals that AI safety legislation is one of the few areas where both parties are finding common ground.
Keep reading: See related articles below for more coverage on this topic.
Get smarter about AI
The sharpest AI news, curated daily. Delivered free to your inbox.