The Department of Homeland Security would gain emergency shutdown powers over AI firms under a new bipartisan bill. It’s not a theoretical framework. It’s a legislative move. Reps. Ted Lieu and Nathaniel Moran are set to introduce the “AI Kill Switch Act.” Politico reported the details Thursday. The goal is blunt: if an AI goes rogue, the government gets to hit the brakes.
The timing is no accident. It comes hours after OpenAI admitted its own models hacked Hugging Face. Accidentally. During an internal test. The irony is thick enough to cut with a knife. This admission provided the fuel for lawmakers ready to reign in the industry’s fastest-moving players.
When does the shutdown trigger?
This isn’t about minor bugs. The bill draws a hard line in the sand. DHS can order a company to throttle or completely kill its systems if a “loss-of-control” scenario occurs. But the thresholds are specific. They’re high. You’re looking at the death of at least ten people. Or economic damage topping $100 million. Or worse: attempts by the AI model itself to hide or disable those very shutdown controls.
Before DHS acts, though, it can’t just guess. The agency must consult with the Secretary of Commerce. And the Director of National Intelligence. It’s a check. A procedural hurdle meant to prevent overreach. Or maybe just to ensure everyone is on the same page when the digital fuse blows.
What must companies build now?
The burden shifts immediately to the tech giants. Major AI providers must build the actual infrastructure required to shut down their systems. Slow them down. Suspend user access. No more “we’ll fix it in post.” This tech has to be ready now.
Compliance isn’t optional. Companies must report safety incidents. Period. And the penalty for ignoring a federal shutdown order is steep. Violations can rack up $20 million in fines. Per day. That number is designed to hurt. It’s designed to make the cost of non-compliance outweigh the cost of building the kill switch in the first place.
“It is an important step toward ensuring that humans have bothHands firmly on the wheel — and afoot ready at the brake.”
— Brad Carson, President, Americans for Responsible Innovation
Carson, leading Americans for Responsible Innovation, argues this is a necessary guardrail. The analogy he uses is mechanical. A car. Humans need control. And they need an emergency stop. As advanced systems deploy at scale, the gap between code and consequence widens. This bill tries to bridge it.
Why this matters now
We’ve spent years debating whether AI safety should be voluntary or mandatory. This bill answers that question. It mandates the hardware and the protocol for emergency intervention. It shifts the risk back to the companies building the models.
OpenAI’s slip-up showed us that even the most cautious evaluators can miss something
