A September executive order directs experts to recommend stronger AI safety laws within two months, including auditors and a possible kill switch for frontier models, as critics warn California is positioning itself as a super regulator for the nation
A New Order From The Capitol
California has long exported its rules to the rest of the country. Its vehicle emissions standards shaped the national car market. Its privacy law influenced how websites treat users everywhere. Now it is moving to set the rules for artificial intelligence. In September, Governor Gavin Newsom issued an executive order convening what his office called world-leading experts and giving them two months to recommend ways to strengthen California’s AI safety and security laws. Reason assistant editor Josh Pauze examined the order in a September 22 piece headlined California wants to write America’s AI rules.
What The Order Does
According to Reason’s account, the order directs the experts to consider requiring frontier AI companies to host independent auditors and follow third-party safety plans, to examine advancing a kill switch for frontier models, and to consider expanding the definition of critical safety incidents to include loss-of-control events. It also accelerates implementation of SB 813 and AB 1405, laws that create a state framework for certifying and registering independent AI auditors, though they do not themselves require developers to submit to audits.
Newsom framed the move as filling a vacuum. With Donald Trump and Congress failing to lead, he said, California is stepping up to keep all Americans safe.
Echoes Of SB 1047
The order revives ideas from SB 1047, the 2024 bill that would have created a state entity to oversee frontier AI models and required mandatory third-party audits, safety testing and kill-switch capabilities. Newsom vetoed it, warning at the time that it could give the public a false sense of security. Two years later, many of the same concepts are back on the table, this time through an executive process rather than a legislative vote.
That shift matters. Legislation is debated, amended and voted on by elected representatives. Executive orders and expert panels move faster and with less public scrutiny. Whatever one thinks of AI risk, decisions that could shape an entire industry deserve the full democratic process.
The Super Regulator Problem
Adam Thierer of the R Street Institute, quoted by Reason, called the order not a guidebook but a rule book, and warned that California is once again attempting to position itself as a super regulator for the entire nation. Because AI systems are developed and deployed across state lines, rules written in Sacramento could effectively become national standards. Companies may find it easier to comply with California’s rules everywhere than to maintain different versions for different states. That is the so-called California effect, and it means 40 million Californians and their legislators can set rules for 330 million Americans who never voted for them.
Thierer also warned that a patchwork of state rules could favour large firms that can afford compliance teams, while burdening smaller developers and start-ups. That is a recurring pattern in regulation: the incumbents that rules are meant to restrain often end up best positioned to handle them, while challengers are squeezed out.
Innovation On The Coast
California’s technology industry grew in a culture of permissionless innovation. Garage start-ups did not need state certification to write code. Surfers turned board-shaping into an industry of independent craftsmen experimenting with new shapes and materials without asking anyone’s leave. That spirit of trying things, learning and iterating built the modern economy of the coast, from surfboards to software.
Heavy-handed AI regulation risks closing that door. A kill switch requirement, for example, raises hard questions: who decides when to flip it, under what standard, and with what accountability? Mandatory auditing regimes create gatekeepers whose approval becomes a condition of doing business. Each step may be justified by safety concerns, but together they can turn a dynamic field into a licensed profession, with the state as the licensing authority.
Supporters of strong AI rules point to genuine concerns, from misinformation and fraud to security incidents and the long-term risks some researchers worry about. Those concerns deserve serious attention. But the answer to uncertainty is not to hand broad discretion to state officials and expert panels.
The Constitutional Question
Reason’s analysis notes that the dormant Commerce Clause, the constitutional principle limiting states from unduly burdening interstate commerce, could in theory restrain state regulation of a national industry. But the Supreme Court has narrowed that doctrine in recent years, and companies have been reluctant to challenge such laws. That leaves the question largely to Congress, which has so far stalled on AI legislation. If Washington does not set national rules, the vacuum will be filled by the most aggressive state, and that state is likely to be California.
A Better Approach
There is a case for targeted rules addressing specific, demonstrated harms, such as fraud, impersonation or the misuse of personal data, enforced through existing laws where possible. There is a much weaker case for broad, pre-emptive control of a technology whose risks and benefits are still being discovered. A sensible approach would let developers innovate while holding them accountable for concrete harms, rather than requiring permission to build in the first place.
The experts Newsom convened will report within two months. Californians, and the rest of the country, should watch closely. The ocean teaches that you cannot control every wave, but you can learn to read them. AI policy built on the same humility would serve the public better than a rule book written in a hurry to claim national leadership.