California issues executive order with major impact on AI development

The December executive order coming from the White House asserting federal authority over AI regulation created a task force to sue states, established a review to catalogue their laws and engineered a mechanism to withhold $42 billion from states that resist the order. Frontier AI is a rare case of technology that Washington has declared a national security priority while simultaneously declining to govern it. That is the gap Governor Gavin Newsom's order on Friday is written into.

Newsom issued Executive Order N-9-26 on Friday, directing the Government Operations Agency to convene national experts and, in consultation with the Governor's Office of Emergency Services, return recommendations within two months on four obligations for frontier AI developers:

• Independent third parties write their safety plans
• Independent verification organizations be embedded onsite in their labs for regular audits
• AI developers must build an emergency shutoff for frontier models
• The state's definition of a reportable critical safety incident expands to cover loss-of-control events

The order gives the agencies until November 15 to report back with advice about what California should require.

Two of these obligations appeared in Senate Bill 1047, the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act, which would have required frontier developers to retain an independent third-party auditor annually beginning January 1, 2026 and to maintain the ability to shut a model down, and which Newsom vetoed on September 29, 2024 on the reasoning that it could restrict development at large companies "at the potential expense of curtailing the very innovation that fuels advancement." California State Senator Scott Wiener, who wrote that bill, now says the state "must act with all possible haste to address serious risks."

Though the policy did not change between 2024 and Friday, the way it will be enforced did. SB 1047 would have made developers liable when a model caused catastrophic harm and they had failed to exercise reasonable care, and it would have allowed the attorney general to enforce that duty in court.

President Trump signed an executive order on December 11, 2025 establishing an AI Litigation Task Force inside the Justice Department to challenge state AI laws as burdens on interstate commerce, directing the Commerce Department to publish a review of state statutes it deems overly burdensome, and authorizing agencies to condition $42 billion in federal broadband funding on states agreeing not to enforce the laws Commerce identifies.

Trump's order identifies California's SB 53 and AB 2013 among its targets, alongside Colorado's AI Act and New York City's Local Law 144. SB 53 is the Transparency in Frontier Artificial Intelligence Act, the 2025 law that created California's frontier-model safety disclosure and critical-incident reporting regime, and AB 2013 requires developers to disclose the data used to train generative models. The first of those is the statute Friday's order proposes to extend, which means the federal government had already painted a bullseye on the law California is now building on.

Colorado as a case study

Colorado is a possible case study of California's future. What happens to a state AI law once a company sues and the Justice Department joins it? xAI sued to block the Colorado AI Act on April 9, the Justice Department intervened on April 24 arguing the Act violated equal protection, and the attorney general suspended enforcement. Three weeks later Governor Jared Polis signed Senate Bill 189, repealing the Act and replacing it with a disclosure-only law, effective January 1, 2027, that drops the duty to mitigate algorithmic discrimination. Colorado never reached court. The statute was rewritten faster than it could be adjudicated.

Read against Colorado, California's choice of an order rather than a bill reads as an effort to delay the moment California has something suable, postponing the part of the legislative process that Colorado could not survive: the existence of a statute. A recommendation commands no one, so there is nothing yet for a court to block. And the two laws the order accelerated, SB 813 and AB 1405, signed September 9, all carry deadlines years out. The Government Operations Agency has until January 1, 2028 to set the criteria for independent verification organizations, and the auditor registry does not open until January 1, 2029, after which no unregistered party may conduct an audit that California law requires. Evidently, California is constructing auditing and regulator bodies on a timeline meant to outlast the administration that currently poses a threat to them.

Whether an order counts as governance, and whether it can do the work of a statute, is the open question we can hope to answer soon. The order's language half-answers it by complaining that "the federal government has done nothing." California, for its part, already has a frontier-model transparency statute in SB 53. What it does not have, and will not have before 2029, is a registry of auditors qualified to certify that anyone has complied with it. Newsom has until September 30 to act on roughly a dozen AI bills still on his desk. Whatever he signs there outlasts his tenure.