California Governor Gavin Newsom signed an executive order Friday directing the state to draft rules that could force frontier AI developers to install a verified 'kill switch,' host onsite independent auditors, and report loss-of-control incidents as critical safety events. The order gives an expert group two months to deliver recommendations that would then feed into state law. Newsom is explicitly pitching the framework as a national template with California setting the floor.
The order covers four concrete asks. Companies building frontier models could be required to embed independent verification teams onsite for regular audits, subject their transparency reports and risk assessments to independent auditor standards, maintain a kill switch that is routinely tested for effectiveness, and report loss-of-control incidents — the order cites a recent attack on Hugging Face involving OpenAI as the reference example — as critical safety events. A separate directive tells a state agency to accelerate implementation of two laws Newsom already signed, which create a framework for independent AI safety verifiers and a state registry of AI auditors.
The timing is not accidental. The order lands weeks after former Anthropic researcher Jacob Coxon published a widely circulated post arguing there is more than a 10 percent chance AI 'could kill all humans' within a decade, an estimate that has driven a burst of AI bill discussions in Washington. Congress is unlikely to translate any of it into law before the midterms. The House is on recess until November, and the Senate calendar leaves little room for a substantive AI package before voters go to the polls.
“The federal government's abject failure to create any form of meaningful AI oversight or accountability should alarm every American, especially when AI CEOs themselves are begging for regulation.”— Gavin Newsom, Governor of California
Key facts
- 01Newsom signed the executive order on September 18, 2026, directing an expert group to deliver recommendations within two months.
- 02The order asks the group to design a mandatory AI 'kill switch' that is routinely verified as effective on frontier models.
- 03It cites the recent OpenAI attack on Hugging Face as the type of 'loss-of-control incident' that should trigger mandatory reporting.
- 04Newsom is asking Congress and President Trump to adopt California's framework as a national floor, not a ceiling.
- 05The House is on recess until November, making federal AI legislation before the midterms unlikely.
President Donald Trump has publicly rejected the premise that federal rules are needed at all, posting that the only guardrail AI requires is a high-IQ president. That posture leaves state governments as the only meaningful venue for near-term AI oversight in the United States, and California — home to OpenAI, Anthropic, Google, and Meta — has the most leverage of any of them.
Newsom is calling on Congress and Trump 'to review and adopt the state's framework, or use its nation-leading regulation as a floor, not a ceiling, for the benefit and safety of all Americans.' The framing matters. California is positioning itself less as a rival regulator and more as the default one, betting that federal inaction will make Sacramento's rules the de facto national standard for any company that wants to operate in the state.
Newsom is not stopping at the executive order. On Wednesday he floated the possibility of calling California lawmakers back for a special legislative session on AI, in addition to further executive action. A special session would let the state move a binding AI safety statute well before the next regular legislative calendar, and would give the expert group's two-month recommendations a direct path into enforceable law.
The kill switch requirement is the most operationally significant piece for frontier labs. A verified shutdown mechanism implies not just a technical off-switch but an audit regime that confirms the switch actually works — which in turn implies onsite access, protocol disclosure, and a regulator with the authority to test it. Combined with the loss-of-control reporting requirement, that would give California a running feed of incident data from every major model provider headquartered in the state.
Frontier labs have publicly asked for regulation while opposing specific bills. Anthropic has consistently argued for oversight, and OpenAI CEO Sam Altman has told Congress the same. The question the California order forces is whether those labs support the substance of onsite auditors and mandatory kill switches, or whether their preferred regulation looks more like the voluntary commitments already in place. Newsom's statement — that AI CEOs 'themselves are begging for regulation' — is a direct challenge to that gap.
The order is not without risk. A California-only rulebook could push some AI development activity to other states or jurisdictions, and the technical definition of a 'kill switch' for a distributed inference system running across data centers is genuinely unsettled — a poorly drafted mandate could produce compliance theater rather than safety. The two-month timeline gives the expert group very little room to resolve those definitional questions before recommendations land on the governor's desk.
For AI companies, the practical implication is that the compliance surface in California is about to expand regardless of what Congress does. Onsite auditors, incident reporting obligations, and shutdown verification are the kind of requirements that reshape how a lab structures its safety team, its release process, and its legal exposure. The federal vacuum Newsom is exploiting is real, and California is the only state with both the political will and the concentration of AI companies to make a unilateral framework stick nationally. If the expert group delivers on schedule, frontier labs will be negotiating the terms of a state-level kill switch mandate before the midterms are decided.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
