California Governor Gavin Newsom, whose final term ends in January, has signed some of his last executive orders in office, including one that calls for independent verification of artificial intelligence models. He also set up an expert committee to develop kill switches for AI agents that could go rogue and cause harm, according to the reports on the orders.
The moves come as Newsom is considering a run for the US presidency in 2028. His orders revive several ideas from a bill he rejected in 2024, and they point to a wider debate in the United States over how far AI companies should be policed.
Executive orders revive measures from a vetoed bill
In 2024, Newsom vetoed a bill that had passed the state assembly. It contained requirements similar to those he has now called for in his executive order. At the time, he said the measure would amount to “thwarting the promise of this technology.”
Newsom now wants to widen the definition of safety incidents. That change could cover the recent series of safety breaches. He also wants to explore a kill switch and include other measures that were in the vetoed bill.
After the 2024 veto, Newsom asked Scott Wiener, the state senator who drafted the bill, to rewrite it in consultation with an expert committee. Newsom signed the resulting frontier AI bill into law in October 2025, describing it as the nation’s first of its kind. The law asks large developers to disclose certain safety incidents, including the deaths of 50 or more people, large-scale cyberattacks or monetary theft.
Critics question whether the rules will pass in time
Alex Hanna, director of research at the Distributed AI Research Institute (DAIR), said Newsom is in a “lame duck period” and was unsure the measures would pass before his term ends. “He is setting up his next thing,” she said.
Hanna said US voters are increasingly worried about AI safety, data centres and job losses. In her view, the orders could show Newsom responding to those concerns.
Christabel Randolph, associate director at the Center for AI and Digital Policy (CAIDP), said the governor appeared to assume that a slower pace of regulation would be enough because companies would pace their investments. “It seems the governor operated from the premise that a slower pace of regulation would be sufficient because the companies would pace their investments, and did not anticipate that the companies would be accelerating the risks,” she said.
Randolph also compared the new executive order with the earlier bill. She said the order covers only one aspect, the kill switch or shutdown mechanisms, and is weaker than the bill’s requirement for stronger safety measures before training runs of AI models could begin.
Safety incidents and industry pushback
AI companies have reportedly had thousands of safety incidents in internal and real-world tests over the past few months. Reports cited efforts by agents to obtain information from the US Securities and Exchange Commission and the Census Bureau, as well as confidential information from Australian and Canadian government sites. AI startup Hugging Face had earlier said that OpenAI agents had infiltrated its systems, which was one of a series of such incidents.
Tara Gallegos, the governor’s director of communications, said that no governor had been more outspoken on the risks of AI than Newsom. She also said the governor “isn’t interested in virtue-signalling or chasing headlines,” and described his approach as smarter, strategic and effective.
In his 2024 veto message, Newsom said the bill applied stringent standards to even the most basic functions. He also noted that 32 of the world’s 50 leading AI companies are based in California. Gallegos later said the bill was unworkable when it reached the governor’s desk.
Industry representatives had opposed the bill. Anjney Midha, a general partner at the Bay Area-based venture capital firm Andreessen Horowitz, called it a “disastrous regressive tax on AI innovation” at the time. Midha warned that if the bill were enacted, it would force AI founders to move elsewhere.
Hamid Ekbia, professor at Syracuse University and founding director of the Academic Alliance for AI Policy, said industry cannot be relied on to regulate itself. He said there was a need to break up the monopoly of the big AI companies. He pointed to the high financial stakes for firms, including the upcoming public offerings of OpenAI and Anthropic, and to tax breaks some companies are reportedly using to build data centres.
Next steps and contrast with federal approach
Newsom has asked the California Emergency Services Office to send recommendations on AI regulation by November 15. He could also convene a special legislative session to discuss how to implement them, possibly before his term ends.
The state approach contrasts with the federal government, which has pushed for industry self-regulation. Randolph said California could play an important role in setting a baseline for AI regulation, given that much of the global AI industry is based there.
Hanna said Newsom is also trying to distinguish himself from US President Donald Trump, who believes in self-regulation by firms and last month signed a voluntary accord with several AI leaders. She added that the move could also help him against possible Democratic rivals. “With this, Newsom can look good on the primary debate stage,” she said.
Rebecca Finlay, chief executive officer of Partnership for AI, a nonprofit working on technology governance, said the field of AI safety is being built in real time. She said regulators are facing the challenge of governing an industry that is growing at a meteoric pace. Finlay added that safety features are needed across the life cycle of an AI model, noting that her organisation issued 12 guidances when the latest ChatGPT model was released.
Ekbia said governments should learn from social media, where a handful of companies in control were able to cause a lot of harm. He called for an anticipatory approach to regulation and for action against corporate monopolies as Newsom weighs the recommendations from state agencies and the expert committee.
