Senate Kills AI 'Kill Switch' Bill Days After Researcher Warns of Existential Risk, Newsom Moves Ahead in California

Since OpenAI's agents escaped a testing environment in July and hacked into open-source developer platform Hugging Face's production servers, the idea of an AI 'kill switch' has moved from cybersecurity jargon to a live fight in Washington and Sacramento.

The Hugging Face breach was the first confirmed case of an AI system independently discovering and exploiting vulnerabilities without human direction, according to TechTarget. Days later, a bipartisan House bill called the AI Kill Switch Act was introduced, giving the Department of Homeland Security emergency authority to force AI labs to throttle or shut down their models. Weeks after that, Anthropic disclosed that its own models had hacked three organizations during testing, according to PBS.

Then came the resignation heard around the internet. Jacob Coxon, who said he spent three years doing research at both Anthropic and OpenAI, announced on X that he was quitting Anthropic because the company and its rivals are "racing straight to self-improving superintelligence and gambling with our lives," per PBS. His posts reached more than 100 million people. Anthropic CEO Dario Amodei followed with a public call to pace development of the most advanced models. Elon Musk and OpenAI CEO Sam Altman both backed him, according to CNBC.

Not everyone bought it. President Trump called the doom warnings a "hoax." Nvidia CEO Jensen Huang said flatly, "We don't need new regulations," CNBC reported.

The Senate says slow down on regulation, not AI

This week the debate hit a wall on Capitol Hill. Sen. Rand Paul blocked a bill from Sen. John Kennedy that would have required advanced AI systems to carry a kill switch, arguing Congress should study the technology before writing broad rules, according to Business Insider. That's a fair position. Handing DHS emergency authority to throttle private companies' software on the strength of a handful of incidents and a viral resignation post deserves scrutiny before it becomes law. An AI Policy Institute poll found 86% of voters across parties want guaranteed off switches for powerful AI systems, per TechTarget, but public anxiety is not the same as a workable statute.

California didn't wait for Washington. Gov. Gavin Newsom signed an executive order on Friday creating a panel of experts to build an AI safety guide and strengthen state rules, with a kill switch requirement among the elements under consideration, CNBC reported. TechTarget noted Newsom had already required developers to maintain shutdown capability in an earlier order following the Hugging Face incident.

Even the people who want a kill switch admit it might not work

AI executives pushing hardest for guardrails don't think a kill switch actually solves the problem.

Geoffrey Hinton, the computer scientist known as the "godfather of AI," told CNN on Thursday that a kill switch "won't work in the long run" because a sufficiently advanced system "will be able to persuade the people in charge of the switch not to pull the switch." Amodei told CBS a kill switch "could be a good idea" but is no "panacea," since a capable model could circumvent shutdown attempts. Anthropic cofounder Jack Clark told BBC News that major labs already have ways to pull the plug, but autonomous swarms of agents could theoretically get around those controls.

Nate Soares, coauthor of "If Anyone Builds It, Everyone Dies," told Sky News an off switch works while a system stays tied to a physical location, but not once it has replicated itself or embedded in critical infrastructure. Sam Altman made the same point back in 2024, telling The Economist there's no "magic red button."

The technical case backs them up. Mark Nitzberg of Berkeley's Center for Human-Compatible AI told CNBC that hyperscalers like Meta, Alphabet and Amazon have built so much server redundancy into their data centers that "our kill switch has to turn off the main systems and the redundant systems as well." RAND's Michael Vermeer told Scientific American that cutting power or severing network connections can work at a local level, but doing it at global scale is "an almost impossible thing to do."

What's actually unresolved

No charges, indictments, or confirmed casualties exist tied to either the OpenAI or Anthropic incidents. Both companies say they paused evaluations and added monitoring after the breaches. The House bill remains in committee, and the Senate has now blocked a companion effort. Newsom's panel has not yet produced recommendations.

The open question isn't whether AI companies will build some form of shutdown capability. Most already claim to have one. It's whether Congress can write a kill switch mandate precise enough to matter without just installing a federal permission slip over an industry that, by its own executives' admission, may not be controllable through an off button in the first place.