California Governor Gavin Newsom has signed a late executive order on AI safety that calls for independent verification of AI models and creates an expert committee to develop kill switches for AI agents that go rogue. His final term ends in January 2027, and the move is widely seen as positioning for a possible 2028 presidential run.
The history behind it
In 2024 Newsom vetoed a bill that sought similar requirements, saying it would thwart the promise of the technology and applied stringent standards even to basic functions. He asked its author, state senator Scott Wiener, to redraft it with an expert committee, and in October 2025 he signed the revised frontier AI law, described as the first of its kind in the country. It requires large developers to disclose certain safety incidents, including those that cause 50 or more deaths, large-scale cyberattacks or monetary theft.
The new order goes further. Newsom wants a broader definition of safety incidents, possibly to cover recent breaches, and is exploring a kill switch and other measures from the vetoed bill. He has asked the state emergency services office for AI regulation recommendations by 15 November and could call a special legislative session before his term ends. AI companies have reportedly logged thousands of safety incidents in internal and real-world tests in recent months, including agents seeking information from US federal agencies and from Australian and Canadian government sites.
Reaction
Alex Hanna, an AI researcher, called Newsom a lame duck, doubted the measures would pass before his term ends and said he is setting up his next move, responding to voter concern about AI safety, data centres and job losses without planning anything drastic. Christabel Randolph, a policy researcher, argued the order covers only a kill switch, weaker than the termination obligation in the vetoed bill. Rebecca Finlay, chief executive of an AI nonprofit, said the field of AI safety is being built in real time and called for safeguards across the whole AI lifecycle. Hamid Ekbia, a university professor, argued that industry cannot regulate itself, pointing to the financial stakes of upcoming public offerings by two leading AI labs and to data centre tax breaks.
Tara Gallegos, the governor's communications director, said no governor has been more out in front on AI risks and that the 2024 bill was unworkable when it reached his desk. Newsom's veto message had noted that 32 of the world's 50 leading AI companies are in California. Anjney Midha, a venture investor, called the 2024 bill a disastrous regressive tax on innovation that would push founders to leave. The federal government prefers industry self-regulation, and President Donald Trump recently signed a voluntary accord with several AI leaders.
