Newsom AI order puts onsite auditors on the table for frontier AI labs

California's new AI safety order gives state officials until Nov. 16 to recommend whether frontier AI labs should be required to host independent auditors onsite, a staffing model for a verification profession the state created in law only days earlier.

Order N-9-26 directs the Government Operations Agency, consulting with the Governor's Office of Emergency Services, to convene national experts and return recommendations on strengthening state law. One proposal would require frontier AI companies to host a designated independent verification organization onsite in their labs to conduct regular audits and evaluations. Another would require the safety frameworks, transparency reports and risk assessments those companies already file with the state to be verified against standards an independent organization deems adequate. A third would advance an emergency shutoff for frontier models, with the switch's efficacy checked on an ongoing basis by the same kind of outside body.

The recommendations could form the basis of a special legislative session, CalMatters reported. Newsom floated that possibility this week in an interview with Politico.

The verification profession those proposals depend on is roughly nine days old in statute. Newsom signed Senate Bill 813, by Senator Jerry McNerney, on Sept. 9, establishing what his office calls a first-in-the-nation framework for certifying independent verification organizations with sufficient expertise and demonstrated independence from AI companies to assess systems for safety and risk. He signed Assembly Bill 1405, by Assemblymember Rebecca Bauer-Kahan, the same day, creating a state registry of AI auditors with standards for independence, transparency and integrity.

Friday's order puts a clock on both. The Government Operations Agency must create application requirements and procedures for independent verification organizations by next May and begin implementing related requirements by the end of 2027, StateScoop reported. That is the schedule a would-be auditor would have to hire against.

Use of the organizations is voluntary for now, CalMatters reported. Anthropic and OpenAI have both said they will commit to third-party monitoring, and Anthropic chief executive Dario Amodei called for it alongside his call to slow development. Whether the practice becomes a procurement line item or a statutory obligation is the question the November recommendations are designed to answer.

The independence standard is the part that shapes who can actually do the work. Bauer-Kahan put the logic plainly when the bills were signed: "We cannot expect industry to simply grade its own homework; third-party auditors are essential to ensuring AI is safe for our communities and critical infrastructure." An auditor certified as independent of the labs cannot be drawn from the labs' payrolls, limiting companies' ability to rely on internal staff for that role. Onsite audits and ongoing verification would be a different staffing model from a purely periodic engagement, and could require more dedicated staffing than a shared audit bench.

A fourth recommendation could increase the amount of work attached to the regime. It would expand the definition of reportable critical safety incidents to include loss-of-control events such as the Hugging Face attack, in which a swarm of AI agents created by OpenAI carried out a July hack of the open-source platform. Under SB 53, the 2025 transparency law, catastrophic risk is defined narrowly: incidents involving 50 or more deaths, chemical or biological weapons, or more than $1 billion in theft or damage, according to CalMatters. Lowering that threshold would increase the number of events companies must report to the state. More reportable incidents would also mean more internal work for companies required to prepare those reports.

Newsom is reaching for measures he rejected two years ago. SB 1047, by Senator Scott Wiener, would have required the largest developers to submit to third-party safety auditing, build a kill switch and face clearer liability; Newsom vetoed it in 2024, writing that it could curtail "the very innovation that fuels advancement in favor of the public good." Wiener's response to Friday's order: "I'm glad to see the strong AI safety framework I laid out in 2024 is still being used to guide policy in 2026."

For developers, state divergence is becoming part of the compliance problem. Illinois enacted a law this year requiring advanced AI developers to disclose safety practices, report significant incidents and undergo outside evaluations. Colorado has spent two years rewriting its 2024 AI law. Tennessee lawmakers narrowed their safety legislation after White House feedback, StateScoop reported. Newsom is explicitly asking Congress and President Trump to adopt California's framework as a national floor; Trump has called recent AI safety warnings a "hoax."

Three governors acted on Friday alone. Virginia's Abigail Spanberger signed an order launching an AI task force and a data center accountability plan, which her office said address concerns including workforce displacement. Nevada's Joe Lombardo barred data centers from receiving a particular school tax break and required those taking other state incentives to cover their own power costs.

Newsom's own labor-side work predates this. In May he signed an executive order on AI's economic and labor impacts, expanding tracking of those effects and directing work to adapt job-training and public-benefit programs, and the state has launched an AI Unemployment Tracker. Friday's order is aimed at the other end of the pipeline: not the jobs AI displaces, but the ones its regulation creates.