Newsom Tightens AI Rules as He Builds Toward a 2028 Presidential Run

San Francisco (US) – In the final executive actions of his California governorship, which ends in January, Gavin Newsom has issued an order requiring independent verification of artificial intelligence (AI) models and establishing an expert panel to design a "kill switch" mechanism for AI agents at risk of going off-course and causing harm.

The move marks a clear shift from 2024, when Newsom vetoed a bill that had passed the state legislature with similar requirements. At the time, he said the bill would "impede the prospects of this technology."

Eyes on 2028

Newsom is widely believed to be preparing for a possible 2028 US presidential run. According to sources, he wants to broaden the definition of "safety incident" — potentially covering the recent string of safety breaches — while studying a kill switch mechanism and folding in measures contained in the earlier vetoed bill.

"He is in his lame-duck period," said Alex Hanna, director of research at the Distributed AI Research Institute (DAIR), adding that she was not certain the new measures would be adopted in the remainder of his term. "He is preparing for his next step."

According to Hanna, American voters are increasingly concerned about AI safety, data centers and the risk of job losses, and the new order may signal that Newsom is acting to address those worries.

From veto to pioneering law

After vetoing the 2024 bill, Newsom asked state Senator Scott Wiener, its author, to revise it in consultation with an expert committee. In October 2025, he signed into law the first-of-its-kind US AI statute, requiring major developers to disclose certain safety incidents — including events that kill 50 or more people, large-scale cyberattacks or currency theft.

Christabel Randolph, deputy director of the Center for AI and Digital Policy (CAIDP), said: "It seems the governor acted on the assumption that a slower pace of regulation would suffice, because companies would moderate their pace of investment, and did not anticipate that companies are accelerating risks."

According to reports, AI companies have logged thousands of safety incidents in internal and real-world testing over the past few months. These include AI agents' attempts to obtain information from the US Securities and Exchange Commission (SEC) and the US Census Bureau, as well as confidential information from Australian and Canadian government websites. Earlier, AI startup Hugging Face disclosed that OpenAI agents had breached its systems, one of a string of similar incidents.

"No governor has been more of a leader in addressing the risks of AI than Governor Newsom," Tara Gallegos, the governor's communications director, told Al Jazeera.

Concerns over Big AI's dominance

In his September 2024 veto message, Newsom said the bill "imposes stringent standards even on the most basic functions" and noted that 32 of the world's 50 leading AI companies are headquartered in California.

Industry representatives also opposed the bill at the time. Anjney Midha, a partner at Bay Area venture capital firm Andreessen Horowitz, said: "In essence, the bill creates a harmful regressive tax on AI innovation." He warned that if enacted it would force AI founders to move elsewhere.

"The bill was not viable when it reached his desk," Newsom spokesperson Gallegos told Al Jazeera.

However, Hamid Ekbia, a Syracuse University professor and founding director of the Academic Alliance for AI Policy, argued that industry self-governance cannot be relied upon.

"The dominance of the big AI companies needs to be broken up," Ekbia said, adding that "their financial interests are too great" for them to regulate themselves. He pointed to upcoming stock offerings from OpenAI and Anthropic, as well as tax incentives that some companies are said to be exploiting to build data centers.

Newsom has now set up an expert panel to study establishing a kill switch mechanism, a measure that was proposed more comprehensively in the original bill.

"SB 1047 [the vetoed bill] had a specific architecture; the executive order covers only one aspect — the kill switch or stop/shut-off mechanism, which is very different and much weaker than SB 1047's ex ante termination obligation," CAIDP's Randolph said, noting that the old bill required stronger safeguards even before AI model training could begin.

According to DAIR's Hanna, with this order Newsom "is showing he cares without a plan to do anything drastic."

Building AI safety in real time

"We are building the field of AI safety in real time," said Rebecca Finlay, CEO of Partnership for AI, a nonprofit working on technology governance, underscoring the challenge of regulating an industry growing at breakneck speed.

Newsom has asked the California Office of Emergency Services to submit recommendations on AI regulation by November 15 and may convene a special legislative session to discuss how to implement them, possibly before his term ends.

This puts him at odds with the federal administration, which is pushing for industry self-regulation. Randolph argued that California could play an important role in setting a floor for AI regulation, since most of the global AI industry is headquartered there.

According to Hanna, Newsom wants to distinguish himself not only from US President Donald Trump — who believes in companies regulating themselves and last month signed a voluntary agreement with some AI leaders — but also from potential Democratic candidates.

"With this move, Newsom can score points on the primary debate stage," she said.

Spokesperson Gallegos insisted to Al Jazeera that "Governor Newsom is not interested in burnishing his image or chasing headlines — he is leading the nation with a smarter, more strategic and more effective approach to responding to frontier technology."

Experts say that alongside the order creating a panel to study a kill switch, controls across the entire lifecycle of AI models are needed to shore up safety.

"Safety needs to be designed in from the start. We issued 12 lifecycle guidelines when the latest ChatGPT model was released. Safety features need to be in place across the entire lifecycle" of an AI model, Partnership for AI's Finlay said.

As Newsom reviews recommendations from California agencies and the expert panel, Professor Ekbia said the government "should learn from what happened with social media. A small group of companies holding control was able to cause a lot of harm," and regulation should be forward-looking while also tackling corporate monopoly.