NAIC Targets AI in Claims Handling at Spring 2026 Meeting: What Reserving Actuaries Need to Prepare

At its Spring 2026 National Meeting in San Diego, the NAIC's Big Data and Artificial Intelligence (H) Working Group singled out AI in claims handling as an area requiring additional regulatory scrutiny. Consumer representatives on the Working Group argued that the lack of transparency in AI-driven coverage decisions and the potential for bias in claims models could violate state unfair claims settlement practices acts, and that claims AI had not received the same level of regulatory attention as pricing and underwriting models. That comment landed at a moment when 88% of auto insurers already use or plan to use AI to evaluate claims, according to NAIC survey data, and when total-loss frequency hit 22.8% through October 2025, on pace for a second consecutive record. From reviewing the NAIC's published AI evaluation questionnaire alongside actual carrier AI governance documentation, the gap between what regulators will ask and what most claims operations can currently produce is significant. This article maps the regulatory timeline, details what the evaluation tool asks that is relevant to claims, and translates the Spring 2026 discussion into actionable preparation steps for reserving actuaries and claims operations leaders.

What the Spring 2026 Meeting Actually Said About Claims

The Big Data and Artificial Intelligence (H) Working Group convened on March 24, 2026, with the agenda split between operationalizing the Model Bulletin and a panel discussion on AI governance trends. The session that matters most for claims professionals was a discussion initiated by consumer representatives on the Working Group who specifically flagged that "the use of AI in claim handling is an area that should be further reviewed." The concern was not abstract. Consumer representatives identified transparency gaps in how carriers use AI to make coverage determinations and flagged the potential for algorithmic bias to produce claim settlement outcomes that would constitute unfair claims settlement practices under existing state law.

The significance of this statement is that it marks the first time the Working Group has explicitly identified claims handling as a regulatory priority distinct from pricing and underwriting. Prior Working Group sessions, dating back through the 2023 Model Bulletin adoption and the 2025 RFI on a potential AI Model Law, treated AI across insurance operations as a single governance challenge. The Spring 2026 meeting drew a line: claims AI raises consumer protection concerns that differ in kind from pricing AI, because the harm from an algorithmically biased coverage denial is immediate and tangible in a way that an opaque pricing factor is not.

Pennsylvania Insurance Commissioner Michael Humphreys, who chairs the Working Group, has framed the broader evaluation effort as a standardized way to surface AI risk before a full market conduct or financial examination is opened. The claims-specific focus adds a sharper edge to that framing. When a market conduct examiner investigates an unfair claims settlement complaint and discovers that the coverage determination was made or materially influenced by an AI system, the examiner will now have a structured framework for asking how the system works, what data it uses, and whether the carrier tested it for bias. That framework did not exist before the evaluation tool, and the Spring 2026 discussion signals regulators intend to use it.

The 88% Problem: AI Penetration in Claims Operations

The NAIC's own survey data quantifies the scale of the regulatory gap. Among 193 auto insurers responding, 88% reported they use, plan to use, or plan to explore AI and machine learning models in their operations. While that figure spans all operational functions, claims is the area where AI has penetrated most visibly. Photo-based damage estimation, automated triage scoring, total-loss threshold determination, fraud detection algorithms, and subrogation opportunity identification are now standard components of the claims technology stack at mid-size and large carriers.

The growth trajectory in claims AI is accelerating. CCC Intelligent Solutions, the dominant claims technology vendor in auto insurance, crossed the 10% AI revenue threshold in Q1 2026 with a $120 million annualized run rate. Between 2024 and 2025, the portion of insurers with full-scale AI adoption expanded from 8% to 34%, though only 22% of carriers that tested AI reached full production deployment. That gap between experimentation and production is itself a governance concern: carriers running AI pilots in claims may lack the documentation infrastructure that full production systems require.

The regulatory timing is not accidental. Repairable claims declined 10.4% through August 2025 compared to the prior year, while total-loss frequency reached 22.8% through October 2025. When AI systems are making or influencing the repairable-versus-total-loss determination for roughly one in five claims, and that ratio is shifting toward total losses, regulators have a concrete consumer protection basis for asking how those determinations are being made.

Claims AI Application	Carrier Adoption Level	Primary Regulatory Concern
Photo-based damage estimation	Widely deployed at top-20 carriers	Accuracy of severity estimates; total-loss threshold bias
Automated claims triage	Growing; 34% full-scale adoption	Routing bias; disproportionate denial rates by demographic
Fraud detection scoring	Mature; deployed across most major carriers	False positive rates; disparate impact on protected classes
Reserve estimation models	Emerging; mostly supplemental to actuarial judgment	Transparency of AI-assisted IBNR; override documentation
Subrogation identification	Moderate adoption	Data lineage; third-party model reliance
Coverage determination support	Early stage; pilot programs	Unfair claims settlement practices; human-in-the-loop requirements

What the Evaluation Tool Asks About Claims

The AI Systems Evaluation Tool, now being piloted across 12 states from March through September 2026, was not designed exclusively for claims. Its four-exhibit structure covers AI across all insurance operations. But several questions within the exhibits map directly onto claims AI systems, and the Spring 2026 claims discussion signals that regulators will use those questions with increasing specificity.

Exhibit A: AI Inventory. The opening exhibit asks carriers to quantify how many AI models are in production, segmented by use case and consumer impact. For claims operations, this means counting every model involved in triage, damage assessment, coverage determination, reserve estimation, fraud detection, and subrogation. Many carriers can identify their high-profile claims AI systems but lack a comprehensive inventory of all models touching the claims workflow, including scoring models embedded in vendor platforms that claims adjusters interact with daily without thinking of them as "AI."

Exhibit B: Governance Framework. This exhibit asks about the AI governance program itself, including roles, responsibilities, vendor oversight, and integration with Enterprise Risk Management. For claims, the governance question has a specific edge: who is accountable when an AI-assisted coverage determination is challenged? If the answer is "the claims adjuster who reviewed the AI output," regulators will ask what training the adjuster received, what information was available to override the AI recommendation, and what percentage of AI recommendations are actually overridden. From tracking early pilot responses, carriers that can demonstrate a documented human override rate alongside the AI recommendation acceptance rate are in a stronger position than those that treat the AI output as effectively final.

Exhibit C: High-Risk Model Details. Claims models that influence coverage determinations, payment amounts, or fraud referrals are likely to be classified as high-risk under most reasonable classification methodologies. Exhibit C asks for development documentation, testing evidence, human-in-the-loop protocols, and compliance review procedures for each high-risk model. The critical question for claims: does the carrier's high-risk classification methodology actually capture claims models, or does it focus narrowly on pricing and underwriting models where regulatory attention has historically concentrated?

Exhibit D: Data Details. The fourth exhibit traces the data feeding AI systems, including external versus internal sources, third-party data licensing, and training data composition. For claims, this means documenting the data lineage behind photo estimation models (whose training images? what vehicle populations? what geographic distribution?), fraud scoring algorithms (what historical claim patterns? what external data vendors?), and reserve estimation models (what loss development triangles inform the training data? how current is the training set?). Exhibit D is the exhibit most likely to be invoked by a market conduct examiner investigating a specific consumer complaint about a claims outcome.

The Vendor Registry Gap: Claims Deferred From Initial Scope

At the same Spring 2026 meeting, the Third-Party Data and Models (H) Working Group refined its proposed vendor registration framework. A significant development for claims operations: the Working Group reached consensus to narrow the framework's initial scope to pricing and underwriting functions only. Claims handling, utilization review, marketing, and fraud detection were explicitly deferred from the first iteration.

This creates an asymmetry. The evaluation tool pilot asks questions about AI across all functions, including claims. But the vendor registry framework, which would give regulators direct visibility into the third-party models that power many claims AI systems, does not initially cover claims vendors. For carriers that rely on vendors like CCC Intelligent Solutions, Tractable, CLARA Analytics, or Verisk for claims AI capabilities, the implication is that regulatory scrutiny of the AI system will land on the carrier through the evaluation tool, while the vendor that built and maintains the model remains outside the registry framework.

The Working Group emphasized that the scope narrowing is iterative, not permanent. Claims handling is expected to be added in subsequent phases. But the gap between the evaluation tool's claims-inclusive scope and the vendor registry's claims-exclusive scope means carriers cannot wait for the registry to solve their third-party claims AI documentation problem. They need contractual audit rights, model documentation provisions, and bias testing evidence from claims AI vendors now, because the evaluation tool questions about vendor oversight (Exhibit B) and data lineage (Exhibit D) apply regardless of whether the vendor is registered.

State Claims AI Laws: The Patchwork Accelerates

The NAIC's Spring 2026 discussion did not happen in isolation. State legislatures have been moving independently on claims-specific AI requirements, creating a patchwork that complicates compliance for multi-state carriers.

Florida HB 527 was the most prominent claims-specific AI bill in 2026. It would have prohibited workers' compensation carriers, insurers, and HMOs from using AI as the sole basis for denying a claim or reducing a claim payment, and required that all such decisions be made by a "qualified human professional" who independently analyzed the claim facts and reviewed the AI output for accuracy. The bill required detailed recordkeeping of the human professional's identity, the date and time of the decision, and the basis for the determination. It passed the Florida House but died in Senate Rules on March 13, 2026. Despite its failure to become law, the bill's requirements represent a template that other states are watching.

Arizona HB 2175 takes effect July 1, 2026, requiring licensed medical directors to personally review and sign health insurance claim denials. While the bill does not specifically reference AI, its practical effect is to mandate human-in-the-loop review for any AI-assisted health claims denial in Arizona.

Colorado SB 24-205, effective June 30, 2026, targets algorithmic discrimination in high-risk AI systems affecting insurance eligibility and coverage decisions. Colorado's approach differs from Florida's in that it does not mandate a specific human-in-the-loop requirement but instead requires developers and deployers of high-risk AI systems to implement risk management practices and conduct impact assessments. For claims AI systems that influence coverage or eligibility determinations, the Colorado framework requires documented testing for algorithmic bias, a standard that goes beyond the NAIC Model Bulletin's general principles.

California, Texas, and Illinois have enacted AI-related requirements touching insurance and healthcare that affect claims workflows indirectly. California AB 3030 requires transparency disclaimers when AI generates clinical information used in coverage decisions. Texas HB 149 and SB 1188 require healthcare practitioners to disclose AI use in diagnostic recommendations that inform coverage determinations.

State	Bill / Regulation	Effective Date	Claims-Specific Requirement
Florida	HB 527	Died March 13, 2026	Qualified human professional for all claim denials; detailed recordkeeping
Arizona	HB 2175	July 1, 2026	Licensed medical director must personally review and sign health claim denials
Colorado	SB 24-205	June 30, 2026	Algorithmic discrimination testing for high-risk AI affecting coverage decisions
California	AB 3030	Enacted	Transparency disclaimers for AI-generated clinical information in coverage decisions
Texas	HB 149 / SB 1188	Enacted	Practitioner disclosure of AI use in diagnostic recommendations

NCOIL's Parallel Track: The Qualified Human Professional Standard

While the NAIC's claims discussion stayed within the boundaries of evaluation and governance, the National Council of Insurance Legislators (NCOIL) has pursued a more prescriptive approach. The NCOIL Model Act Regarding Insurers' Use of Artificial Intelligence, introduced at the July 2025 meeting and sponsored by New York Assemblyman Erik Dilan and Oklahoma Representative Forrest Bennett, would have required a "qualified human professional" to make the final decision on all insurance claims.

The NCOIL model act mandated that before denying a claim or reducing a payment, the qualified human professional must independently analyze the claim facts and policy terms, review the AI-generated outputs for accuracy, and review previous human-made decisions on the claim. Insurers would be required to maintain records of all actions taken by the qualified human professional and explain to the claimant how the AI system was used.

Industry opposition was strong. ACLI, NAMIC, APCIA, and AHIP submitted a joint letter arguing the one-size-fits-all approach was inappropriate and that existing technology-neutral laws already covered AI-assisted claims decisions. Assemblyman Dilan acknowledged the consensus challenge in early 2026, and development of the model was paused to allow the legislative landscape across states to evolve. NCOIL plans to resume discussions at its November 2026 annual meeting, which coincides with the NAIC Fall National Meeting where the evaluation tool is expected to be finalized.

The timing matters for claims operations leaders. If NCOIL resumes its model act development and the NAIC simultaneously adopts the evaluation tool with enhanced claims-specific questions, carriers could face two parallel frameworks: one prescriptive (NCOIL's human-in-the-loop mandate), the other governance-focused (NAIC's documentation and transparency requirements). Multi-state carriers will need to meet the more demanding standard in any state that adopts one or both.

What This Means for Reserving Actuaries

The claims-specific regulatory focus creates obligations that extend beyond the claims department and into actuarial reserving functions. When AI systems influence claim severity estimates, triage decisions, or settlement timing, those outputs flow into the data that actuaries use to set case reserves and estimate IBNR. If a regulator questions the AI system producing those outputs, the actuarial opinion that relied on the downstream data is implicated as well.

From reviewing the evaluation tool exhibits alongside ASOP No. 43 (Property/Casualty Unpaid Claim Estimates) and ASOP No. 56 (Modeling), several specific preparation steps emerge for reserving actuaries.

Document the AI contribution to reserve data. If claim severity estimates, claim closure rates, or settlement amounts are influenced by AI systems, the reserving actuary needs to understand and document how those AI outputs enter the reserving data pipeline. This is not about validating the AI model itself; it is about knowing whether the loss development triangles being used in reserve projections reflect AI-influenced settlement patterns that differ from historical patterns. When an AI triage system routes straightforward claims to automated settlement while flagging complex claims for manual review, the resulting development patterns will differ from patterns generated when human adjusters handled both categories.

Establish override documentation protocols. Model validation for rate filings already requires documentation of how AI recommendations are accepted, modified, or overridden. Reserving actuaries need an analogous framework for claims AI. What is the override rate on AI-generated severity estimates? How often do adjusters increase or decrease an AI-recommended reserve? A regulator asking about bias in a claims AI system will want to see whether the override pattern itself shows demographic variation. If adjusters override AI recommendations more frequently for certain claim types or claimant profiles, that pattern is a regulatory risk that the reserving actuary should be aware of even if the AI model itself tests cleanly.

Test for AI-driven shifts in development patterns. When AI systems accelerate claim settlement for certain categories while slowing others, the resulting loss development factors will reflect those shifts. An actuary selecting development factors from a period before AI deployment and applying them to a period after deployment may be selecting factors that systematically misstate ultimate losses. The reverse is also true: factors calculated from AI-influenced experience may not be appropriate for projecting losses in states or lines where AI is not deployed. This is a standard actuarial judgment call under ASOP No. 43, but the AI context means the judgment must be documented with explicit reference to the AI system's influence on the underlying data.

Prepare for regulatory questions about AI-assisted reserve estimates. If a carrier uses AI to generate initial case reserve estimates that are then reviewed by adjusters or actuaries, the evaluation tool's Exhibit C questions about high-risk model details apply directly. Regulators will want to know how the model was developed, what data it was trained on, how often it is retrained, what testing was performed for bias, and what the human override protocol is. The NAIC's March 2026 AI Issue Brief states clearly that "existing state insurance laws apply regardless of whether" decisions involve humans or algorithms. A reserve estimate generated by an AI system is held to the same standard as one generated by an actuary.

Compliance Timeline: Key Milestones Through November 2026

The regulatory calendar is compressed. From the Spring 2026 meeting through the expected adoption of the finalized evaluation tool at the Fall 2026 meeting, carriers have roughly six months to address claims-specific governance gaps.

Date	Milestone	Action Required
March 2, 2026	AI evaluation tool pilot begins across 12 states	Claims AI inventory and governance documentation should be underway
March 24, 2026	Working Group flags claims as priority area	Claims departments should review evaluation tool exhibits for claims-relevant questions
May 11, 2026	Deadline for pilot participants to submit information to Working Group	Pilot state carriers should have completed initial Exhibit A responses
June 30, 2026	Colorado SB 24-205 takes effect	Carriers writing in Colorado must have algorithmic discrimination testing for claims AI
July 1, 2026	Arizona HB 2175 takes effect	Health carriers in Arizona must implement medical director review for AI-assisted denials
September 2026	Evaluation tool pilot ends	Pilot feedback informs tool revision; carriers should have full claims AI governance documentation complete
September-October 2026	Tool revision and public re-exposure	Review updated tool for new claims-specific questions; submit comments if warranted
November 2026	Expected adoption at Fall National Meeting; NCOIL resumes model act discussions	Finalized compliance posture for adopted tool; monitor NCOIL model act trajectory

Practical Steps: Building Claims AI Governance Before the Window Closes

Whether or not a carrier is in one of the 12 pilot states, the evaluation tool sets the standard for what regulators will expect. From watching how carriers in pilot states are responding and comparing that against what the evaluation tool exhibits actually ask, five concrete steps stand out.

First, build a complete claims AI inventory. Count every model in the claims workflow: triage scoring, damage estimation, coverage determination support, fraud detection, reserve estimation, subrogation identification, litigation propensity scoring, and settlement recommendation. Include vendor-provided models that claims adjusters interact with through existing platforms. Many carriers discover during this exercise that their claims AI footprint is larger than their governance framework covers.

Second, classify claims AI systems by risk level. Apply the carrier's existing high-risk classification methodology to the claims AI inventory. Any model that influences coverage determinations, payment amounts, or fraud referrals should be classified as high-risk. If the existing methodology does not capture claims models, revise it. The NAIC's four-tier risk taxonomy (unacceptable, high, medium, low) provides a reference framework, and the evaluation tool lets carriers set their own criteria, but regulators will challenge boundaries that exclude claims systems with direct consumer impact.

Third, document human-in-the-loop protocols for every high-risk claims AI system. For each system, record what decision the AI recommends, what information the human reviewer sees, what authority the reviewer has to override the recommendation, and what happens when an override occurs. Track override rates and review them for demographic patterns. The NCOIL model act's "qualified human professional" standard may not become law in 2026, but the concept of documented human oversight is already embedded in the evaluation tool's Exhibit C.

Fourth, secure vendor documentation rights. For third-party claims AI systems, review vendor contracts for provisions covering model documentation access, bias testing results, training data composition, and regulatory cooperation obligations. The vendor registry framework may not cover claims initially, but the evaluation tool asks about vendor oversight regardless of registry status. Carriers that cannot produce documentation about how a vendor's claims AI model was developed and tested will face a gap that the vendor's registry status does not close.

Fifth, coordinate between claims, actuarial, and compliance functions. Patterns we have seen in early pilot responses confirm that carriers designating a single coordinator for evaluation tool responses handle the process more cleanly than those distributing it across departments. Claims AI governance touches claims operations, actuarial reserving, compliance, legal, and IT. A cross-functional team with clear ownership produces more consistent documentation than parallel efforts that are reconciled after the fact.

Where This Is Headed

The NAIC's claims-specific focus at the Spring 2026 meeting is not an endpoint. It is a signal that claims AI will receive the same level of regulatory scrutiny that pricing and underwriting AI have attracted over the past three years. The evaluation tool pilot, running through September 2026, is the mechanism regulators are using to build competency and gather evidence about how claims AI operates in practice. The results from the 12 pilot states will inform whether the finalized tool includes claims-specific exhibits or more targeted questions, and whether the Working Group's claims discussion leads to formal guidance or a broader push toward a model law that treats claims AI distinctly.

The March 2026 NAIC AI Issue Brief reinforced that "existing state insurance laws apply regardless of whether" decisions involve humans or algorithms. For claims operations, that principle means the Unfair Claims Settlement Practices Act, which exists in some form in every state, applies to AI-driven claims decisions with the same force it applies to decisions made by human adjusters. The evaluation tool does not create new legal obligations; it creates a structured way for regulators to assess compliance with obligations that have always existed.

For reserving actuaries, the practical takeaway is that claims data influenced by AI systems requires the same level of documentation and professional judgment that has always been required under ASOP No. 43 and ASOP No. 56, but with an additional layer of awareness about how the AI system's behavior shapes the data. The regulatory window between now and November 2026 is the time to close that documentation gap, before the evaluation tool moves from pilot to standard practice.