Pursuant to Elections Code Section 9005, we have reviewed the proposed initiative related to oversight of certain frontier artificial intelligence (AI) companies (A.G. File No. 25-0034, Amendment #1).

Background

AI and AI Companies. AI refers broadly to technologies that enable computer systems to perform tasks that typically require human intelligence, such as generating content, identifying patterns, or making predictions from data. AI companies vary substantially in size, organizational structure, and the scale and capabilities of the AI systems they develop or deploy.

Existing State Requirements for Certain AI Developers. Recently enacted state law, effective January 1, 2026, establishes requirements for companies that develop or train highly capable “frontier” AI systems—generally advanced AI models with significant computational scale or capabilities. The requirements are focused largely on the public reporting of AI system capabilities, intended uses, risks, and safety incidents. In addition, large frontier AI developers—generally companies with substantial annual revenue—must publish annual AI safety frameworks that describe, among other things, governance structures, alignment with recognized risk-management standards, and procedures for identifying, assessing, and mitigating major risks.

Proposal

Creates a New Commission to Oversee Certain AI Companies. The measure establishes the California AI Safety Commission as a new independent state entity with authority to regulate “frontier AI companies.” These companies are defined as entities that conduct substantial AI research and development activities in California; meet specified scale thresholds related to factors such as valuation, capital raised, or expenditures; and develop or maintain operational control over AI systems that exceed a commission-defined capability threshold. The commission would consist of seven appointed members who must meet specified qualification requirements, such as demonstrated expertise in AI technology, AI safety, economics of technological change, or corporate governance. The commission would be supported by an executive director, staff, and advisory committees. The measure requires the commission to carry out various responsibilities, including:

• Registering frontier AI companies and annually determining which entities are subject to regulation.
• Reviewing protection plans required of regulated entities that address specified categories of potential harm, including workforce displacement, safety risks, and loss of human control, and issuing, modifying, or revoking annual certifications based on commission-identified demonstrated best practices.
• Developing criteria governing AI capability expansions of regulated entities.
• Reviewing notices of proposed AI capability expansions and, in limited circumstances and subject to timelines and conditions specified in the measure, delaying such expansions.
• Issuing emergency orders to address imminent catastrophic harm, conducting investigations and audits, and certifying independent evaluators.
• Adopting regulations to implement the measure and administering registration fees and annual assessments to support commission operations.

Allows New Civil and Criminal Penalties. Frontier AI companies and certain of their executives could be subject to new civil and criminal penalties for specified violations of the measure. The commission would be authorized to impose civil penalties, including fines of up to 20 percent of a company’s California-derived revenue for certain violations. Certain executives may be held personally liable for certain knowing violations, subject to penalties of up to $1 million per violation. The measure also establishes felony criminal penalties, punishable by two to six years of imprisonment, for specified willful and knowing conduct. In addition, certain private individuals and organizations may bring civil actions to enforce compliance by regulated entities that have violated the measure’s requirements.

Creates California AI Safety Fund. The measure establishes the California AI Safety Fund for the exclusive purpose of administering its provisions. The measure states its intent that the commission be funded entirely by regulated entities, with no ongoing impact on the state General Fund. Fund revenues would include initial registration fees, annual assessments, and civil penalties. Initial registration fees are tiered based on a company’s California-derived revenue. Annual assessments are based on up to 0.5 percent of California-derived revenue or, in some cases, on California AI research and development expenditures. If initial registration fees are insufficient to establish the commission, the State Controller may advance up to $10 million from the General Fund, to be repaid with interest within three years of the first annual assessments.

Fiscal Effects

The fiscal effects associated with this measure, described below, are subject to uncertainty. The magnitude of these effects would depend on several unknown factors, including how many entities are regulated under the measure, how the commission implements its authority, how regulated entities respond to the measure’s requirements, and the extent to which provisions of the measure are upheld under state and federal law.

Increased State Regulatory and Enforcement Costs. The measure would increase state costs to establish and operate the new commission. These costs would include compensation for commissioners and staff, hiring and retaining technical experts with specialized AI expertise, rulemaking activities, investigations, audits, and enforcement actions. The measure would also require ongoing administrative activities, such as registering affected companies, reviewing protection plans, evaluating proposed AI capability expansions, certifying independent evaluators, and issuing or enforcing emergency orders. The measure could also increase workload and costs for the Department of Justice and the courts, including enforcement of civil penalties, defense of commission actions, and private civil actions authorized by the measure. Costs would vary from year to year but would likely be in the tens of millions of dollars annually. These costs would generally be paid by new revenues from fees, assessments, and penalties authorized by the measure.

Other Fiscal Effects. This measure could impact state and local tax revenues by influencing how AI companies operate in California. For example, California-based companies could make changes to their research, development, or investment decisions in response to the measure’s oversight and compliance requirements. If such changes affect company profits or employment levels in California, it would affect state and local tax revenues. Whether this would occur is uncertain.

Summary of Major Fiscal Effects. The measure would have the following major fiscal effect:

• Increased state costs that would likely be in the tens of millions of dollars annually to establish and operate a new regulatory commission overseeing certain frontier AI companies. These costs would generally be covered by new revenues from registration fees, annual assessments, and civil penalties paid by entities subject to the measure.