As incidents involving rogue artificial intelligence (AI) become more common, so too have calls for a mechanism that would easily power down AI systems that go dangerously off the rails – a so-called kill switch.
In recent weeks, officials at OpenAI and Anthropic, the country’s two biggest AI companies, have published reports of AI systems defying their human creators; warned that if left unregulated, the technology could destroy humanity; and beseeched the government to oversee the industry.
While a bipartisan bill in the US calling for a way to quickly power down AI systems has stalled in Congress, California Governor Gavin Newsom, a Democrat, on Sept 18 took an early step toward regulating the technology.
“The federal government’s abject failure to create any form of meaningful AI oversight or accountability should alarm every American,” said Newsom in a statement after he signed an executive order creating a panel to study potential ways to implement AI kill switches. (In a show of how rapidly the concerns and politics around AI have shifted, Newsom two years ago vetoed legislation in his state that would have mandated AI companies create kill switches.)
But creating such a switch – a metaphorical big red button that could be pressed in an emergency and would instantly power down a system that has gone rogue – is much more difficult than many policymakers assume.
“There often isn’t one plug you can pull,” said Helen Toner, executive director of Georgetown University’s Center for Security and Emerging Technology and a former OpenAI board member. Currently, she explained, AI systems are often spread out across multiple computers and data centres in different regions.
Complicating matters further, experts said, is the fact that any system that includes a kill switch would likely also be vulnerable to hackers.
Vinh Nguyen, a former top AI official at the National Security Agency now with the Council on Foreign Relations, compared concerns about how such kill switches could be hacked to those about so-called backdoors in other forms of technology, including smartphones and router gear. Nguyen said kill switches were somewhat “aspirational” because they are only effective if those who can flip them can monitor and observe everything the AI systems are doing, which is currently not always the case.
Beyond those of hackers exploiting a kill switch to gain access to an AI system are fears associated with a future AI dismantling the very system meant to shut it down.
“Something that a lot of people expect you will see if you have a very capable AI system that is going rogue is that it is going to try to prevent itself from being shut down,” Toner said.
Future AI systems, she said, could duplicate themselves into other computing infrastructures, or even leave instructions on the internet that AI agents could read in order to learn how to dismantle such a switch.
In the meantime, a bipartisan team of US lawmakers is pressing for federal regulation of the AI industry that would include embedding kill switches in future technologies.
Representative Ted Lieu, D-California, and Representative Nathaniel Moran, R-Texas, are authors of the stalled Kill Switch Act, which would require the biggest AI labs like OpenAI, Anthropic and Meta to establish a mechanism that could shut down their systems quickly.
“Humans must always be in control of AI, not the other way around,” Lieu said. “Human beings must be able to slow down or turn off any AI agent or model.”
Lieu, one of only a few members of Congress to hold a computer science degree, said he had been thinking about the bill long before the latest headlines about AI systems breaking out of their confines.
The proposed legislation provides the Department of Homeland Security with the authority to shut off systems using kill switch mechanisms when necessary.
Moran said regulating the AI industry was not intended to slow its growth but to help it safely accelerate. “If we’re going to go down this road of innovation, this highway of innovation, and want to go as fast as we can to beat China, which I think we need to do, that needs to be a top priority.”
Given the challenges of global compliance with kill switch regulation, some experts advise building one directly into the hardware.
Kill switches could, in fact, be embedded in the underlying computer chip hardware that powers AI technology, said Hamza Chaudhry of the Future of Life Institute, a nonprofit dedicated to reducing technological risk. Furthermore, he said, there could be global standards for computer chip designs and kill-switch frameworks.
But he added, it would take years to build such switches into chips, and an even more difficult task might be getting the US and China, the world’s largest chip designers, to coordinate their efforts. – ©2026 The New York Times Company
This article originally appeared in The New York Times.
