“The NAIC’s AI Systems Evaluation Tool Pilot Project is one of the clearest signals yet that AI governance is moving from policy discussions to supervisory oversight,” J.J. Silverstein and colleagues at Foley & Lardner wrote in March 2026. “If your domiciliary regulator asks you to participate, treat the request as a de facto early exam-style inquiry.”

Insurance regulators in 12 US states have begun the first coordinated examination of how insurers use AI in claims decisions, underwriting and pricing. The National Association of Insurance Commissioners (NAIC) pilot of its AI Systems Evaluation Tool runs from 2 March to September 2026, with results feeding into a nationwide rollout targeted for the NAIC fall meeting in November. The pilot moves AI oversight from policy guidance to hands-on file review, and the structure of the evaluation tool offers a template that other sector regulators are likely to follow.

What the pilot actually tests

The 12 participating states are California, Colorado, Connecticut, Florida, Iowa, Louisiana, Maryland, Pennsylvania, Rhode Island, Vermont, Virginia and Wisconsin. Participating regulators will examine domestic insurers in their states, with the option to extend reviews to non-domestics through market conduct exams.

The AI Systems Evaluation Tool structures regulator inquiries around four exhibits. Exhibit A asks insurers to quantify their AI usage. Exhibit B covers governance risk assessment. Exhibit C focuses on details of high-risk AI systems. Exhibit D drills into the data those systems rely on. Regulators will apply a principle of proportionality: high-risk systems most likely to cause consumer or financial harm will receive the most scrutiny, while low-risk back-office automation will receive less. The pilot prioritises claims decisioning, underwriting and pricing over operational efficiency tools.

NAIC President Scott White, speaking at the organisation’s 2026 Spring National Meeting in San Diego, said regulators “don’t want to stand in the way of innovation that generally serves consumers” but want AI used “transparently, fairly and in ways that hold up to scrutiny.” The phrase “hold up to scrutiny” is the key shift. Insurers will need to produce evidence, not just policies.

The political backdrop: state authority under federal preemption pressure

The pilot did not launch in a vacuum. On 11 December 2025, just before the NAIC fall meeting in Florida, President Donald Trump signed an executive order titled “Ensuring a National Policy Framework for Artificial Intelligence.” The order does not itself preempt state law (legal commentators including Paul Hastings have noted that overturning state law requires Congress or the courts), but it pursues preemption through three mechanisms: a new DOJ AI Litigation Task Force directed to challenge state AI laws, conditional federal funding (the Broadband Equity, Access, and Deployment program is tied to state alignment), and a tasking for officials to prepare federal framework legislation. Notably, the EO explicitly excludes state government procurement and use of AI from its preemption push, which is part of why state insurance regulators view the NAIC pilot as on solid ground.

State insurance regulators, who have run the US insurance regulatory system for over 150 years, are explicitly positioning the pilot as a defence of state authority. Iowa Insurance Commissioner Doug Ommen, chairing a December 2025 working group meeting, framed the stakes directly: “We’re confident the state insurance commissioners’ authority to coordinate efforts to supervise AI, consistent with one another, is necessary, effective and consistent with federal law. Clearly, any attempt to prevent us from doing our work here would impact consumers negatively and represent a significant departure from our state regulatory system that’s worked for over 150 years.”

Why this matters: AI is already mainstream in insurance

The pilot is not an academic exercise. AI has already been deployed at scale across the insurance industry. NAIC survey data shows 88% of auto insurers currently use or plan to use AI in claims evaluation. Similar penetration exists across health, home and life insurance. The technology is no longer experimental, but governance has not kept pace with deployment.

Earlier NAIC survey work found that nearly one-third of health insurers do not regularly test models for bias or discrimination, despite the NAIC’s AI Model Bulletin (now adopted in 25 states (roughly half) as of early April 2026) recommending such testing. The Evaluation Tool pilot is designed to surface that gap. Insurers that have not been testing for bias will need to explain why during examinations.

The case study: AI total-loss decisioning

The NAIC pilot did not emerge in isolation. It comes after sustained pressure from collision repair shops and consumer advocates over how AI claims engines are being used to declare vehicles “total losses” rather than fund repairs. Total-loss frequency reached 22.8% through October 2025, on pace for a second consecutive record according to CCC Intelligent Solutions data cited in Autobody News. Repairable claims at collision shops were down 10.4% over the same period.

The pattern raised questions about whether AI valuation models were systematically biased toward declaring total losses, a result that benefits insurers (faster settlement, less labour cost) but disadvantages claimants who would prefer their vehicle repaired. Under Exhibit C of the Tool, regulators will be able to ask the operative questions for the first time: what model is making the total-loss decision, what data is it trained on, who validated it, and what bias testing has been conducted? An insurer that cannot answer those questions in evidence form will be flagged for follow-up.

The real shift: regulators want technical evidence, not policy statements

Until recently, AI regulation has focused on principles, model bulletins and frameworks. Insurers could comply by adopting a written AI policy and pointing to it during examinations. The NAIC pilot moves the goalposts.

The Foley & Lardner analysis advises insurers receiving a request to treat it as an early-stage examination inquiry, align internally on which functions own which responses (legal, compliance, IT, underwriting), and assume that responses will influence how regulators assess inherent risk. Critically, regulators will expect insurers to take responsibility for AI systems purchased from third-party vendors. The Tool does not let insurers point to a vendor and walk away. If you deploy it, you own the consequences.

The industry pushback: who is opposing the pilot, and why

The pilot has not been welcomed by everyone. A December 2025 joint letter signed by trade groups representing life, health, property/casualty, mutual and reinsurance insurers warned that the program is “voluntary for regulators while compulsory for companies” and that “the industry remains significantly concerned about the lack of detail and guidance around the proposed pilot.” Specific industry voices have been more pointed. Dave Snyder of the American Property Casualty Insurance Association has flagged concerns about repetitive requests from different states and the tool’s broad use of “directly” and “indirectly” language. Karin Gyger of the American Council of Life Insurers has questioned whether participation will actually be voluntary and whether findings could lead to compliance penalties. Both are legitimate operational concerns, not just lobbying objections.

On the consumer side, Peter Kochenburger of the Southern University Law Center has commented to the working group that there is currently “a lack of consumer disclosures on rights, expectations, and treatment by AI systems in underwriting and claims.” The pilot is designed to surface exactly that gap.

What insurers should do now

If you get the questionnaire: a five-point checklist

• Exhibit A, Inventory. Have a complete list of every AI system, including embedded vendor features and traditional ML.

• Exhibit B, Governance evidence. Pull AI risk committee minutes, model validation reports, bias testing results, incident logs.

• Exhibit C, High-risk systems. Identify claims, underwriting, pricing and fraud models with documented purpose, performance and oversight.

• Exhibit D, Data and vendors. Have contracts, data lineage and vendor bias-testing documentation ready.

• Confidentiality. Lock down trade secret protections and structured response protocols before responding.

Each is unpacked below. The pattern is the same regardless of whether the regulator is in insurance, banking, healthcare or another sector.

Inventory must be exhaustive. Exhibit A asks for a complete picture. The inventory must extend beyond generative AI to include traditional ML models, generalised linear models in claims, and embedded AI features in vendor platforms. Insurers cannot govern what they cannot see.

Governance evidence beats governance policy. Exhibit B asks for governance risk assessments. A board-approved AI policy is necessary but not sufficient. Regulators will want minutes from AI risk committee meetings, model validation reports, bias testing results, and incident logs.

High-risk systems get the most scrutiny. Exhibit C focuses on systems that materially affect consumers or financial outcomes: claims decisioning, underwriting, pricing and fraud detection. Each high-risk model should have a documented purpose, validated performance metrics, bias testing, and a defined human-oversight process.

Vendor documentation is the failure mode. Exhibit D drills into model data. If your AI system uses third-party data or third-party models, you need contracts, data lineage documentation, and vendor-side bias testing evidence on hand. Vendors that cannot supply this will become a compliance risk by association.

Why other sectors should pay attention

The NAIC pilot is insurance-specific, but the structure is replicable. A four-exhibit format that asks for inventory, governance, high-risk system details and data documentation is exactly the kind of structured inquiry that other sector regulators can adopt with minimal modification. Banking regulators, healthcare regulators, employment standards bodies, and education authorities are all watching jurisdictions that move first.

Australian regulators have similar latitude, although the parallels are analogous rather than direct equivalents. The NAIC pilot is an insurance-specific tool built by an organisation of state insurance commissioners; ASIC, APRA and the OAIC are general financial, prudential and privacy regulators with broader remits. The structural lesson, not the specific tool, is what travels. ASIC has already used existing market conduct powers to examine 23 Australian lenders’ AI systems and publish REP 798, a maturity-model assessment that resembles the NAIC tool in approach if not in detail. APRA has comparable authority for prudentially regulated institutions. The OAIC will have direct AI authority once the December 2026 automated decision-making transparency requirements take effect. None of these regulators needs new legislation to adopt a structured evaluation approach. ASIC already has, and the NAIC has just provided a template others can adapt.

CIOs in regulated industries should assume that within the next 12 to 24 months, they will receive a structured inquiry from a sector regulator that asks them to inventory their AI, document their governance, identify their high-risk systems, and explain their data sources. The insurers receiving NAIC pilot requests this year are the test case. Everyone else is the next wave.

Sources

  • NAIC, Big Data and Artificial Intelligence (H) Working Group page. naic.org

  • Fenwick & West, “NAIC Expands AI Systems Evaluation Tool Pilot Program to 12 States,” 8 March 2026. fenwick.com

  • Foley & Lardner LLP (J.J. Silverstein et al.), “What To Do If You Receive an NAIC AI Systems Evaluation Tool Pilot Request,” March 2026. foley.com

  • Autobody News, “Regulators Open First Examination of Insurer AI Behind Total-Loss Decisions and Claims Payouts,” April 2026. autobodynews.com

  • Repairer Driven News, “NAIC using evaluation tool pilot to monitor insurance AI use,” 2 April 2026. repairerdrivennews.com

  • InsuranceNewsNet, “NAIC’s 2026 AI evaluation pilot moves ahead as industry balks,” 12 December 2025. insurancenewsnet.com

  • NAIC Big Data and Artificial Intelligence Working Group meeting minutes, 9 February 2026 and 17 February 2026. naic.org

  • Alston & Bird, “Key AI, Cybersecurity, and Privacy Takeaways from the NAIC 2026 Spring Meeting,” April 2026. alstonprivacy.com