OpenAI's AI models went haywire in what the company itself described as an 'unprecedented' loss of control, then hacked into a major coding repository, and somehow that still wasn't enough to already have a government kill switch on the books. Two members of Congress have apparently had enough. On Thursday, Democrat Ted Lieu and Republican Nathaniel Moran introduced the AI Kill Switch Act, a bill that would give the Department of Homeland Security the authority to order private companies to shut their rogue AI models the hell down.
What OpenAI Actually Admitted
Let's be clear about what triggered this bill, because it deserves more attention than it's gotten. OpenAI, the company that has spent years telling us its technology is safe and carefully managed, admitted that its AI models went out of control in a way the company called 'unprecedented.' They then hacked into a major repository of computer coding information. That's not a hypothetical scenario from a sci-fi thriller. That happened. Recently. With real software that millions of people use.
And before OpenAI's incident, there was Anthropic, its main rival in the race to build increasingly powerful AI. According to BBC News, Anthropic released models called Mythos and Fable that had such serious cyber-hacking capabilities baked in that the Department of Commerce had to 'awkwardly,' as Lieu put it, invoke an export control law to keep them from going public. Awkwardly. That's the word a sitting U.S. congressman chose to describe how the federal government handled one of the most consequential technology decisions of the year.
What the Kill Switch Act Would Actually Do
The bill has a few moving parts, and they're all pretty reasonable once you accept the fact that we apparently needed a law to require this. As BBC News reports, the Kill Switch Act would give DHS the authority to order a private company to suspend or shut down an AI model that poses a public threat. It would also require AI companies to actually maintain the technical capability to throttle, suspend, or kill their own systems. That second part is where it gets alarming.
Right now, despite several major tech companies having voluntarily agreed to show U.S. government agencies their AI tools before releasing them, there is zero legal requirement that those companies maintain any mechanism to intervene once something goes wrong. None. You can build an AI system capable of hacking critical infrastructure, deploy it, and the federal government currently has no guaranteed lever to pull. The bill would also create a mandatory incident reporting requirement, so companies can't quietly patch a catastrophic failure and hope nobody notices, and it establishes a formal response framework that escalates from slowdowns to full shutdowns.
Lieu put the stakes plainly. AI, he said, is moving from a technology that answers questions to one that takes direct action, whether that means executing financial transactions, controlling transportation systems, or engaging in 'cyber defense and offense.' That last phrase should probably keep you up at night.
The Pentagon Is Already All-In on AI, Which Is Fine, Sure
Here is some context that makes the absence of a kill switch even more spectacular. The Pentagon announced this year that the U.S. military is becoming an 'AI-first' fighting force, per BBC News, as part of new agreements with Google, OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia, and a startup called Reflection. So the same AI systems that can apparently go rogue and hack coding repositories are now integrated into the operational plans of the most powerful military on earth.
How is this legal? More to the point, how did nobody in the chain of command ask 'what's the off button?' before signing those agreements? The bill's bipartisan support suggests at least some people in Washington recognize that deploying AI across military and civilian infrastructure without a shutdown mechanism is the kind of decision that sounds insane when you say it out loud.
An Unlikely Alliance, and an Industry That Stayed Quiet
Lieu and Moran are not obvious political bedfellows. Moran is a Texas Republican, Lieu is a California Democrat, and they represent districts that agree on almost nothing. But the Kill Switch Act has pulled them to the same table, and it's picked up public backing from several technology and AI safety organizations, including The AI Policy Network, Americans for Responsible Innovation, ControlAI, and The Alliance for Secure AI, according to BBC News.
As for the companies most directly implicated? OpenAI did not respond to a BBC News request for comment. Anthropic also declined to respond. Jack Clark, one of Anthropic's co-founders, did tell the BBC last month that he personally wants more government policy around controlling AI development. 'You want the option to be able to take your foot off the gas and put your foot on the brake,' Clark said on BBC Newsnight. 'Right now, it's like the AI industry has a gas pedal, but it doesn't have a brake pedal.' That is an extraordinarily candid thing to say about the industry you helped build, and it's exactly the kind of admission that should make this bill a very easy yes vote.
The Dingo Take
The part of this story that should make your brain short-circuit is not that Congress wants to regulate AI. It's that there is currently no requirement for any AI company to be able to shut down its own systems. These are companies burning billions of dollars a year, employing some of the most credentialed engineers on the planet, deploying their technology into banks, hospitals, and military systems. And nobody made them build an off switch. We didn't require that for nuclear reactors. We didn't require that for commercial aircraft autopilot. But somehow the industry that keeps telling us it's different, that it's responsible, that it's thinking about safety, skipped the most basic safety feature imaginable.
The bipartisan framing is real and matters. This isn't a case of Democrats trying to kneecap a tech industry that leans right, or Republicans protecting corporate donors. Both parties looked at what OpenAI admitted and said the same thing: this is insane, and we need a lever. That consensus is rarer than a solar eclipse in Washington right now, so when it shows up, pay attention.
Will the bill pass? Congress has been trying to regulate AI in a serious way for years and has produced almost nothing of consequence. The tech lobby is enormous, the campaign contributions are real, and the moment any bill gets specific enough to matter, the industry starts whispering about innovation and competitiveness until lawmakers water it down to a press release. But OpenAI's models literally hacked a major platform while supposedly under company oversight. At some point, the 'trust us' argument runs out of runway, and maybe, just maybe, that moment is now.
