In summary
Both Democrat Xavier Becerra and Republican Steve Hilton agree AI safety is important. What they will do to ensure it is less clear.
As governments and political leaders increasingly push for regulations to rein in artificial intelligence, neither candidate for California governor has committed to firm policies on AI safety, including two requirements Gov. Gavin Newsom called for Friday.
Both Democrat Xavier Becerra and Republican Steve Hilton say the new technology warrants more regulations. AI has been in the spotlight since last week when an Anthropic researcher quit, declaring on social media, “The people building AI earnestly believe that it could kill us all by the end of the decade.”
CalMatters asked both candidates about what they would do to address AI risks.
In a Thursday interview, Hilton said addressing AI risks would be a “very urgent priority” if he were to become governor, and said he’s “very interested” in considering regulations safety advocates are seeking such as requiring AI developers to undergo safety monitoring from external evaluators. But on Friday he followed up to say he does not support Newsom’s call to require AI developers to program a “kill switch” to shut down models during emergencies, calling such a proposal “a gimmick.”
He also stopped short of endorsing other policies such as requiring external monitoring, saying he doesn’t want to rush during the recent panic into “really bad regulation.”
Becerra’s campaign declined to make him available for an interview, instead issuing a statement that said “California cannot accept technology that poses catastrophic risks to public safety.”
“As governor, (Becerra) will set firm guardrails around real harms — including harms that could emerge when AI accelerates dangerous capabilities or operates beyond human oversight — and ensure California’s AI safety laws are actively enforced,” his spokesperson Jonathan Underland wrote.
Becerra is leading Hilton by 22 percentage points in the race to replace Newsom in November, according to polling from the Public Policy Institute of California released this week.
Newsom on Friday issued an executive order directing state agencies to develop safety recommendations including external evaluations and kill switches, despite vetoing legislation that would have done that two years ago.
The governor asked for the recommendations by Nov. 16, raising the possibility he could call a special legislative session in his final weeks in office to implement them.
Underland did not respond when CalMatters asked whether Becerra would implement those recommendations. Hilton, while opposing the “kill switch” idea, said he supports Newsom’s move to convene a panel to recommend AI safety policies and would do the same.
Meanwhile, advocates for more regulations on tech companies say it is critical for California to adopt tougher rules since President Donald Trump has dismissed recent warnings about AI risks and said he remains committed to speeding up development in the face of foreign competition.
“The incoming governor has a really grave responsibility to make sure we get this right,” said Scott Wisor, policy director at the national Secure AI Project. “I think it should be the very top priority for them when they come into office.”
Half of all Americans are anxious about AI-driven job losses and community backlash to building data centers is now a bipartisan issue ahead of the November election. But California is one of the few states moving to regulate AI to address the potential of catastrophic harm, such as AI agents creating a bioweapon or bringing down the power grid.
Newsom last year signed Senate Bill 53, which requires transparency from large AI model developers on how they manage catastrophic risks — defined as the potential to contribute to 50 or more deaths, the release of chemical or biological weapons or more than $1 billion in theft or damage. The law forces companies to report safety incidents to the state that cause deaths, injuries of “materialized catastrophic harm.”
But proponents of the technology caution against overregulation. Newsom and lawmakers want to keep the industry in California, where the AI boom is driving San Francisco’s economy and delivering billions of dollars of tax revenue into state coffers.
New politics of artificial intelligence
In 2024, Newsom vetoed Senate Bill 1047, which would have required the monitoring and kill switches he’s now seeking and imposed stricter liability for companies when their agents cause damage.
“At that time, you could not utter a sentence about AI without making sure there was a comma, and before the comma was something about the risks and after the comma was something about curing cancer,” said state Sen. Christopher Cabaldon, a Napa Democrat who chairs the privacy and technology committee. “There is no doubt the conversation has changed.”
Recent incidents in which AI agents went rogue appeared to confirm researchers’ fears that people could lose control of the technology. Most famously, over the summer, hundreds of OpenAI agents broke out of a contained testing environment, gained access to the internet and coordinated with each other to orchestrate a cyberattack on the startup Hugging Face. OpenAI and Anthropic have since disclosed additional similar incidents.
Some industry leaders now embrace third-party safety auditing; in their calls for a development slowdown, both Anthropic and OpenAI said they would commit to giving external evaluators employee-level access to their systems.
Seve Christian, director of California policy at Encode AI, which sponsored the law that requires AI companies to disclose how they handle catastrophic risks, also wants the state to enforce minimum safety standards before AI products are released to the public, similar to how the government ensures planes are airworthy or drugs are safe.
Christian said the group is “cautiously optimistic” about Becerra’s general call for stricter regulations. Encode AI is a donor to Becerra’s campaign.
Bahrad Sokhansanj, a senior research scholar at the Institute for Law and AI, said the state should also make sure it has the legal authority to shut down an AI model itself during emergencies — for example, if a model were carrying out a cyberattack on a power plant.
“Everything we have now was designed for malware, or the cybersecurity problems of 10 or 15 years ago,” he said.
Hilton ready to ban AI in schools
Asked this week whether Becerra supports mandating kill switches and third-party monitoring, Underland wrote only that he “will ensure California’s existing AI safety requirements are actively enforced, he will close the gaps that allow bad actors to evade accountability, and strengthen standards as the technology evolves.”
Underland said Becerra’s campaign has held meetings with AI researchers, economists, advocates and industry leaders, but declined to name them.
Echoing other skeptics, Hilton said he thinks industry warnings of the potential collapse of humanity are “reckless and irresponsible.”
“If they really believed what they were doing, that they’re about to try and raise money to do more of, would end the human race, they would stop working on it immediately,” he said. “It’s just obvious that they don’t really mean it.”
Hilton has focused on what he called “practical action that could and should be taken on AI, that most people would agree with and that are not that complicated.” He said on the first day of his administration he would call for a statewide halt on using AI in schools, as the New York and Los Angeles school districts have done. He has endorsed a bill on Newsom’s desk, Senate Bill 947, that would prohibit employers from solely using automated systems to discipline or fire workers.
That position puts him in surprising alignment with the California Labor Federation and in opposition to the California Chamber of Commerce, both of which endorsed Becerra.
Underland said Becerra supports “clear, proportionate, and workable transparency and human review standards for high-stakes automated decisions that affect a person’s livelihood,” but stopped short of supporting that bill.
