California pushes AI kill switch concept, accelerates oversight for frontier models

California Governor Gavin Newsom signed an executive order that pushes forward a "kill switch" framework for frontier AI models and accelerates the state's safety oversight timeline from 2029 to 2027. For teams building or deploying advanced AI systems in California, this signals that formal safety assessments and takedown mechanisms could become a compliance requirement sooner than expected.

Newsom signs executive order to accelerate AI safety oversight and explore kill switch

The executive order, signed on September 18, 2026, calls on the federal government to take a more active role in regulating AI, while directing California state agencies to speed up implementation of existing AI safety laws. Newsom criticized the Trump administration for what he called abdicating "its responsibility to protect Americans" from AI risks. The order requires companies developing frontier models to develop a "kill switch" in the event models go rogue, though the details are still being worked out.

The order builds on recently enacted California legislation that establishes a framework for independent organizations to assess AI systems for safety and potential risk. It also accelerates the AI auditor registry and oversight program from January 2029 to December 2027.

What a kill switch would actually mean for frontier models

Newsom acknowledged that the kill switch concept is still largely in its infancy and will be refined by a panel of experts he plans to convene. A kill switch could be a technical mechanism to deactivate or constrain a model if it behaves dangerously, but the exact implementation will depend on the expert panel's recommendations. The order directs exploration rather than imposing immediate technical requirements.

The move follows a failed federal effort: Senator John Kennedy introduced a federal AI kill switch bill that was blocked by Senator Rand Paul, who argued against regulating the entire industry without input from leaders.

Accelerated timeline: from 2029 to 2027

Originally, California's AI auditor registry and oversight program were set to launch in 2029. The executive order moves that target to December 2027, cutting nearly two years off the timeline. This acceleration applies to the safety assessment framework and the registry, meaning developers should expect state-level oversight mechanisms to become operational within about 15 months.

What this means for AI builders in California

The order signals that California is moving toward enforceable governance for frontier models. Developers and companies deploying advanced AI systems in the state may need to:

• Prepare for independent safety assessments through the AI auditor registry.
• Design models with potential takedown or kill switch capabilities in mind, even if the technical specifications are not yet finalized.
• Monitor the expert panel's recommendations, which will shape any eventual requirements.

Newsom's order also explicitly calls for a more active federal role, suggesting potential coordination between state and national approaches. However, the current federal administration opposes new AI regulations, creating uncertainty about how state and federal rules will interact.

Caveats and uncertainties

The order directs exploration and framework development, not immediate mandates. The kill switch concept remains undefined, and the expert panel has not yet convened. Compliance requirements will depend on the final framework, which could take months or years to develop. Additionally, political opposition at the federal level could complicate enforcement or lead to legal challenges.