Newsom Plan Would Give California Power to Shut Down Frontier AI Models
California would get licensed AI verifiers, 48-hour incident reporting and standby caps on recursive self-improvement under Newsom's November 18, 2026 legislative recommendations.

Why it matters
- Newsom released the legislative recommendations on November 18, 2026, based on a Government Operations Agency report dated November 16, 2026.
- Emergency shutdown orders would take effect immediately and remain in force during judicial review.
- Most critical incident reporting deadlines would drop from 15 days to 48 hours; 24 hours for imminent catastrophic risks.
- Capability-based reporting would cover models under 10^26 operations if they can create catastrophic risk.
- Recursive self-improvement would face standby cap authority rather than an immediate limit.
California Governor Gavin Newsom on November 18, 2026 released legislative recommendations that would let state officials order immediate emergency shutdowns of frontier AI models posing imminent catastrophic risks — with those orders staying in force during judicial review.
The shutdown powers are the sharpest edge of a broad proposal from the California Government Operations Agency. The report, titled AI Safety in 2026 and Beyond and dated November 16, 2026, is intended to inform legislation when lawmakers return in December 2026. Its recommendations are not yet law. If enacted, they would move California beyond its current requirement that large frontier developers publish safety frameworks, and toward continuous, state-supervised scrutiny of the industry's most powerful systems.
Who would verify AI safety — and how?
The centerpiece is a system of licensed independent verification organizations, or IVOs. Every large frontier developer would face ongoing oversight by these licensed assessors rather than one-time safety filings.
The IVOs would have real teeth. They would translate the state's standards into technical tests and update them as AI evaluation methods evolve. They would get access to models, personnel, systems, training environments and records they judge necessary. Their assessments would go to the state unredacted.
The Government Operations Agency would license the IVOs and could revoke licenses for insufficient expertise, independence or rigor. If an assessor could not attest that a developer met the standards, the state could order corrective action after engaging with the company. Developers could challenge such determinations in court before they take effect — except under the expedited emergency process.
There is a funding mechanism attached. A surcharge on the verification fees developers pay would help finance publicly operated CalCompute capacity. Some of that compute would be reserved for independent safety and security research and public oversight, tying an industry-funded fee directly to the infrastructure external scrutiny requires.
How would an emergency shutdown actually work?
Each large frontier developer would maintain a Risk Containment Protocol, tested and verified on an ongoing basis by an IVO and certified as effective by a senior company official. The required capability would span compute, agents, internal access and model deployment — not a single on-off switch.
The protocols would identify responsible personnel, coordinate with cloud providers and power suppliers, and establish when a model can safely be restored. Testing would address the ways shutdowns can fail in practice, including a model copying its own weights to obstruct containment or an outside actor seizing control of shutdown systems.
Open-weight developers get a carve-out that reflects technical reality. Containment and shutdown requirements would apply only to deployments under their control, because they cannot shut down copies hosted by others.
State officials could order an immediate shutdown when they have credible reason to believe there is an imminent catastrophic risk or major loss-of-control event — and the developer refuses to contain the danger. The order would require the narrowest workable intervention, take effect immediately and remain in force during judicial review. Developers could still seek a temporary restraining order. The order would be lifted once the danger is no longer imminent.
What changes in incident reporting?
The reporting regime would expand on three fronts:
- Attempted model-weight theft would be reportable even without death or bodily injury.
- All internal or external loss-of-control incidents would fall under the regime.
- A new near-miss reporting channel would be created.
The deadlines tighten considerably. The report proposes cutting the deadline for most critical incident reports from 15 days to 48 hours, while keeping the existing 24-hour requirement for imminent catastrophic risks. Whistleblower protections would extend to former employees and contractors, closing a gap that often silences people after they leave a company.
The thresholds shift too. Models capable of actions that could create catastrophic risk would be treated as frontier models for incident reporting even when trained with fewer than 10^26 integer or floating-point operations. That capability-based trigger addresses the report's stated concern: relying on training scale alone has left dangerous incidents outside state reporting requirements.
What about AI building AI?
The report takes automated AI research seriously as a regulatory target. Companies automating AI research and development would report quarterly on how their compute is divided among training, research experiments, inference for AI-research agents, safety and security work, external customers and unattributed activity. IVOs would verify the categories.
Additional disclosures would cover:
- training-data volume, cost, subject domains and human-versus-AI provenance;
- inference compute used to train models in reinforcement-learning environments;
- rewarded misbehavior and the mitigations applied.
The report says these measures would show whether AI is taking over more of its own development and whether safety investment is keeping pace.
Recursive self-improvement, or RSI — whose strongest form involves an AI system independently developing its successor without human input — would face standby cap authority rather than an immediate limit. Officials could impose a time-limited, periodically reviewed cap on automated AI research when it exceeds a developer's committed thresholds or advances faster than its safeguards and evaluations, creating a material risk of catastrophic harm. The cap would be lifted after an IVO verified adequate safeguards were in place.
Why this landed now
The recommendations answer Newsom's September 18, 2026 executive order on independent oversight and an AI kill switch, which requested expert proposals within two months and accelerated implementation of California's independent-verification and AI-auditor laws. The Government Operations Agency developed the report with the California Office of Emergency Services and four outside advisers: Jason Goldman, Gillian Hadfield, Alondra Nelson and Rob Reich.
For developers, the stakes are operational, not cosmetic. Compliance would extend beyond publishing safety policies to continuous external scrutiny, faster incident reporting and demonstrable shutdown capabilities. That is a different order of burden from the disclosure regimes most frontier labs already navigate.
The proposal leaves lawmakers with two connected decisions: how much intervention authority to enact, and how to make continuous oversight operational while the licensed verification sector is still developing. How Sacramento answers both will shape whether the most aggressive state AI oversight regime in the US can actually function.
Source: unite.ai
More from Elena Vasquez
Show full bio
Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.
744 articles
Related articles
- California proposes mandatory independent AI risk reviews
- California Governor Signs Laws Shielding Workers From AI Threats
- Khanna Bill Would Ban Self-Improving AI Until Federal Safeguards Exist
- OpenAI's public policy agenda: safety rules, youth protections
- OpenAI Publishes Frontier Governance Framework for AI Risk