SACRAMENTO, Calif. — California Gov. Gavin Newsom has ordered state officials to evaluate tougher artificial intelligence safeguards, including independently verified “kill switches” for advanced AI models and on-site audits of major AI developers.
Newsom’s executive order, signed Friday, directs the California Government Operations Agency to submit recommendations by Nov. 16 on potential amendments to the state’s existing AI safety and security laws. The agency is to consult with the Governor’s Office of Emergency Services and national experts.
The proposed safeguards the agency must evaluate include:
- Requiring large frontier AI developers to allow designated independent verification organizations to operate inside their laboratories and conduct periodic audits and evaluations.
- Requiring independent verification of the safety frameworks, transparency reports and risk assessments that frontier AI companies must file.
- Requiring frontier AI models to have a “kill switch” whose effectiveness would be tested continually by an independent organization.
- Expanding the types of safety incidents AI companies must report to include a broader range of cases in which developers lose control of their systems.
No Immediate Requirements
The order does not immediately impose those requirements on AI companies. Instead, it directs state officials to examine their technical feasibility and potential effectiveness and recommend possible changes to California law.
Newsom cited reports that people have attempted to use AI products to develop biological weapons and that AI agents have defeated security protections and conducted hacking operations, in some cases going undetected for months.
The governor said those developments raise serious public-safety concerns and require California to act in the absence of federal legislation. He also said the state must balance tighter protections with continued technological innovation and economic growth.
Under the order, the Government Operations Agency must publish application requirements, procedures and standards for organizations seeking certification as independent AI verifiers by May 1, 2027. The agency also faces a Dec. 1, 2027, deadline to carry out additional requirements under California law governing AI safety oversight.
Newsom said California previously established standards for state procurement and use of AI, developed training requirements, examined the technology’s effect on vulnerable communities and launched pilot programs intended to improve public services. The state has also enacted legislation establishing baseline safety and transparency requirements for frontier AI developers.
The executive order took effect immediately.
