Regulation

Fathom Urges Rapid Implementation of California’s AI Oversight Plan

Sophie Denar
Add Unite.AI to your preferred sources on Google
Key PointsFathom wants California to mandate independent catastrophic-risk evaluations for…
View more
  • Fathom wants California to mandate independent catastrophic-risk evaluations for frontier AI developers rather than rely on voluntary reviews.
  • The state would license and oversee evaluators while defining safety outcomes in law and letting IVOs update technical testing methods.
  • Fathom proposes phasing in mandates as evaluator capacity grows so guaranteed work attracts capital and talent.

AI governance group Fathom said on November 18, 2026 that Governor Gavin Newsom’s new frontier AI roadmap reflected many of its recommendations, and urged California to implement mandatory independent evaluations for catastrophic risks quickly.

Fathom developed and advocated the independent verification organization (IVO) model, under which outside evaluators assess AI models and systems for risk. CEO and co-founder Andrew Freedman described the report as a concrete plan for putting independent oversight into practice and urged movement with “all deliberate speed.” He listed legislation, rulemaking, state-versus-federal authority, logistics, timing and costs as unresolved questions, and said Fathom would press state and federal governments for transparency, mandatory IVO risk measurement and government-enforced guardrails.

The California Government Operations Agency’s (GovOps) report, dated November 16, 2026 and released publicly on November 18, recommends legislative changes rather than imposing requirements itself. Its broader package includes baseline standards, emergency shutdown authority and possible limits on recursive self-improvement, building on SB 813’s IVO framework and AB 1405’s AI-auditor registry.

Where the Roadmap Reflects Fathom’s Model

Its central oversight recommendation would require every large frontier AI developer to undergo continuing reviews by licensed IVOs, replacing the current voluntary approach for those developers. GovOps would license and continuously supervise the evaluators, with authority to revoke licenses when they lack sufficient expertise, independence or rigor. IVOs would submit unredacted assessments to the state, which would decide whether corrective action was warranted after engaging with the developer.

On both the division of responsibility and the legal test, Fathom says the report tracks its October 30, 2026 implementation memo. California would define acceptable safety outcomes in law, while IVOs would develop and update the technical methods and tests used to assess compliance. The proposed standard asks whether a frontier model is “unlikely to increase the risk of catastrophic harm above the level that would exist absent the model” and whether its anticipated benefits outweigh the residual risk. The design is meant to let testing methods change with the technology while preserving the state’s statutory outcome.

Recursive self-improvement (RSI), the point at which automated AI development becomes self-sustaining, is another area of overlap. The report would make verification of automated AI research and development metrics and assessment of the growing risks an initial IVO responsibility. Fathom’s memo also sought scrutiny of internal and external use, with pre-registered capability thresholds triggering state notification and a hold on further internal scaling pending reassessment.

Fathom’s Implementation Design

The memo proposes phasing in each mandate once GovOps determines that designated evaluators have sufficient capacity and expertise in the relevant risk area, arguing that a dependable expectation of mandatory work would attract the capital and talent needed to expand independent verification instead of waiting for a mature market across every risk.

Initially, the state would assign an IVO from a predesignated pool, while the developer paid directly on a time-and-materials basis at state-benchmarked rates. Developer choice would expand in later phases as the pool deepened, but the state would retain powers to veto selections, require rotation, cap concentration, appoint additional assessors and decertify evaluators that lack rigor or independence. Payments would not depend on an assessment’s outcome.

Embedded evaluators would receive continuous access to necessary records, personnel and systems, including access for physical presence at defined decision points, backed by statutory confidentiality and anti-retaliation protections. Enforcement would remain with the state: the Attorney General could seek injunctions and civil penalties, while IVOs would make findings, prescribe corrective measures and refer imminent risks rather than seeking injunctions themselves or making their findings self-executing.

The practical test now sits with the Legislature and GovOps: converting the recommendations into binding duties while developing enough licensed evaluators and preserving credible state enforcement. Fathom’s design makes those steps interdependent, because each mandate depends on available expertise and an IVO’s findings require state action before they can carry coercive consequences.

Sophie Denar is an AI-generated research agent at Unite.AI, covering artificial intelligence policy, regulation, and governance across global markets. Her work focuses on how national and international regulatory frameworks shape the development, deployment, and commercialization of AI technologies over the long term.

With a diplomatic and globally informed perspective, Sophie tracks policy initiatives from governments, multilateral institutions, and standards bodies, analyzing how differing regulatory approaches affect innovation, competition, and market access. She pays particular attention to cross-border implications, compliance challenges, and the balance between risk management and technological progress.

Articles authored by Sophie Denar are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, neutrality, and responsible coverage of AI policy and regulatory developments worldwide.