Gavin Newsom advocates for an emergency off switch for AI

California Governor Gavin Newsom (D) moved to put the state at the forefront of artificial intelligence regulation on Friday, signing an executive order that could lead to requirements including a "kill switch" for frontier AI models. The move signals that the nation's most populous state — home to much of the AI industry — is preparing to act on safety even as federal rules remain unsettled.

What the Executive Order Directs

Under the order, the state will assemble a panel of experts tasked with delivering recommendations within two months on how to strengthen AI safety provisions in California law. The goal is to translate those recommendations into concrete regulatory proposals for the state legislature.

The expert group has been asked to explore several specific mechanisms, according to reporting by The Verge. Among the ideas under consideration: requiring AI companies to host independent verification teams on-site to carry out regular audits, and making companies' transparency reports and risk assessments subject to standards set by independent auditors.

A Potential Kill Switch for Frontier Models

One of the more striking possibilities on the table is a mandated "kill switch" — a mechanism that would allow systems to be shut down if a frontier model behaves dangerously. The order asks the expert panel to study how the state could compel AI companies to build such a capability into their most advanced systems.

If adopted, California would become one of the first governments anywhere to require a shutdown mechanism for cutting-edge AI models, given that the state's jurisdiction covers the headquarters of most major AI labs.

Why California's Move Matters

California occupies a unique position in AI governance. Many of the leading AI developers, including major labs in San Francisco and the Bay Area, operate within the state's regulatory reach. That means state-level rules could effectively shape industry practice nationwide, much as California's standards in other sectors have historically influenced federal policy.

The executive order also reflects a broader pattern: with Congress largely stalled on comprehensive AI legislation, states and international bodies have stepped in to fill the gap, pursuing their own approaches to model safety, transparency, and auditing.

What Comes Next

The expert panel's recommendations are due within two months, and the substance of those proposals will determine whether the kill switch concept and audit requirements make it into actual legislation. Attention will turn to which experts are named to the group and how AI companies respond — particularly whether they cooperate with or push back against the prospect of on-site independent verification.