NAIC Creates Market Conduct Working Group to Examine AI-Era Insurers

The NAIC's Market Regulation and Consumer Affairs (D) Committee appointed a Market Conduct Regulation Modernization (D) Working Group on March 25, 2026 in San Diego, with recommendations due by the Fall 2026 National Meeting.

The distinction that makes it consequential is jurisdictional. Every other NAIC AI body is writing rules for carriers. This one is examining whether the examination process can check them.

Key Takeaways

March 25, 2026 formation with a Fall 2026 deadline gives the group roughly seven months to produce recommendations on data collection, interstate collaboration, the Market Regulation Handbook, NAIC systems and examiner training.
23 states plus Washington, D.C. have adopted the December 2023 AI Model Bulletin, with Colorado, Connecticut and Texas layering their own statutes on top of it.
12 states are running the AI Systems Evaluation Tool pilot from March through September 2026, and its results land at the same Fall meeting as the modernization recommendations.
Four exhibits structure that pilot: AI system inventory, governance frameworks, high-risk model details, and data inputs and lineage. Exhibit A is where carriers have had the most difficulty.
The (D) Committee, not the (H) Committee, owns this work, which is why it addresses examination method rather than AI standards.

What the Five Charges Cover

Market conduct regulation has never had the structural overhaul solvency received after 2008, when the Solvency Modernization Initiative produced ORSA, group supervision and risk-based capital reform. It has run on incremental updates to the Handbook and the Market Conduct Annual Statement.

The charges name five gaps. Data collection: whether MCAS fields, designed around human-processed applications and manual claims review, capture anything about a gradient-boosted triage model or language-model correspondence. Interstate collaboration: coordination of both analysis and examination. The Handbook and examination approaches, built for paper-file review and claims sampling. NAIC systems capacity. And examiner training.

Illinois Director Ann Gillespie, who chairs the parent committee, described the aim as a comprehensive assessment identifying opportunities to enhance policyholder protection "while creating efficiencies for regulators and companies." NAIC President and Virginia Commissioner Scott A. White framed the question as whether the current framework "has the data, the system, the tools, and the supervisory approaches needed to oversee a rapidly evolving industry."

Rules Without a Verification Path

The reason this group is separate from the others is that the NAIC's AI work has been building substance without building the mechanism that checks it.

The Big Data and Artificial Intelligence (H) Working Group produced the Evaluation Tool, runs the 12-state pilot, proposed the four-tier risk taxonomy, and is operationalizing the Model Bulletin. The Third-Party Data and Models (H) Working Group is building a vendor registry, having narrowed Phase One from six insurance functions to pricing and underwriting. Both define what carriers must do.

Neither reaches how a market conduct examiner verifies it. The Handbook does not specify what documentation to request when a decision was made by a model, so an examiner can arrive with a complete AI governance standard and no method for testing compliance against it. That gap is the working group's entire remit.

The operational consequence lands on documentation rather than on modelling technique. Exhibit A of the pilot asked carriers for a comprehensive inventory of AI and machine learning models in production, and it is the exhibit that exposed the widest gap, because most carriers hold model records across risk, IT and business units rather than in one defensible list.

If standardized MCAS fields follow, that inventory becomes a filed quantity: model counts, deployment contexts, consumer-facing applications, third-party vendor usage. A validation file built for internal review is not the same artifact as one built to be produced on request across 23 states plus D.C., and the second is what a filed inventory implies.

Exam timing could move with it. Scheduling currently runs off MCAS analysis, complaint trends and periodic cycles, all of which lag. Tying review to AI deployment events instead, a new automated claims system, a third-party underwriting vendor, a material expansion of algorithmic decision-making, would put examination frequency on the same clock as the change being examined rather than a year behind it.

Coordination Requires States to Accept Each Other's Work

The charge with the most reach is interstate collaboration, and it is the one that runs into something the NAIC cannot resolve by drafting.

A carrier running one underwriting model across 40 states can face examinations in a dozen jurisdictions applying different standards, requesting different documentation and reaching different conclusions about the same model. Colorado's AI Act imposes quantitative bias testing beyond most other states, Connecticut has its own governance framework, Texas passed TRAIGA with distinct disclosure requirements, and the 12-state pilot adds a further layer, as counsel summaries of the Spring meeting set out. Four compliance regimes for one model is the current arithmetic.

Examiner capability is the other half. Examining an AI-driven process asks for model governance, algorithmic fairness testing, data lineage and vendor oversight, which is a different skill set from reviewing claims files, and the NAIC's own AI issue brief frames technical competency as central to state oversight. Competency standards for examiners are within the training charge, and they raise the sophistication of the questions actuarial teams will be answering.

The obvious fix is a lead-state examination with multi-state recognition, on the pattern financial examination follows under the accreditation program. Market conduct has historically resisted exactly that, because states treat consumer protection as a sovereignty matter in a way they do not treat solvency review. A framework the group proposes in November is therefore a voluntary instrument, and its reach depends on adoption rather than on drafting quality.

The calendar compresses this further. The pilot is expected to produce results in September through October 2026, with tool revisions and re-exposure before adoption at the Fall meeting, and the modernization group's initial recommendations are due at that same meeting. Findings about where the Evaluation Tool worked and where it did not will arrive alongside, rather than ahead of, the recommendations they are meant to inform. The group's output is labelled initial for that reason, which places the substantive coordination question in 2027 rather than at the deadline being worked to now.

NAIC Creates Market Conduct Working Group to Examine AI-Era Insurers

Key Takeaways

What the Five Charges Cover

Rules Without a Verification Path

Coordination Requires States to Accept Each Other's Work

Further Reading

Sources