NAIC Plans to Consider an Updated AI Evaluation Tool for Fall Adoption

The National Association of Insurance Commissioners plans to consider the AI Systems Evaluation Tool for adoption in November, at the 2026 Fall National Meeting, according to the NAIC’s pilot project summary. That same working timeline puts one step first. Across September and October the instrument is to be updated on pilot feedback and issued again for review, so what would reach the fall meeting is the updated version rather than the one pilot states are using now. The tool, which has been running as a 12-state pilot since March 2026, is designed to give state examiners a common framework for assessing how insurers govern AI systems.

The timing matters. If the tool is adopted in the fall, it becomes part of the examiner toolkit for 2027 market conduct, financial analysis, and financial condition exams. That moves AI governance from a forward-looking project to a 2027 exam topic. The questions are about counts, risk assessment, high-risk system documentation, and where the data came from, and they have to be answered out of the file that exists on the day the examiner asks.

The tool’s four-exhibit structure is already public: counts of AI models by operational area (A), a governance risk assessment (B), detail on the systems the company itself calls high risk (C), and the categories of data feeding those systems with each source named (D). The pilot is testing whether those questions produce useful evidence and whether carriers can answer them with documents they already have. The feedback from pilot states will shape the updated version, but the core structure is unlikely to change dramatically.

Adoption in the fall would not mean every state uses the tool immediately. The exhibits are optional by design, and the NAIC tells regulators to trim them to the scope of a given inquiry rather than send the whole set. State adoption of NAIC instruments is uneven, and some states may take time to integrate the tool into their examination processes. However, the largest states and the most active market-conduct regulators tend to follow NAIC guidance quickly. Carriers operating in multiple states should assume that, if the tool is adopted, at least a subset of regulators will begin using it in 2027, and that others will use it as a benchmark even if they do not formally adopt it.

The practical work for carriers between now and the fall meeting is to pressure-test their own documentation against the four exhibits. A good test is to ask whether a regulator could walk from Exhibit A to Exhibit D in your company without finding contradictions. If the inventory says a system is low-risk but the governance committee spent three meetings on it, the mismatch is a problem. If the sources listed for a high-risk model’s data do not match the model documentation, the examiner will notice.

The summer is also the window to refresh AI governance training. Examiners using the tool will put their questions to business-line owners as well as compliance officers. If the person who runs the claims system cannot explain how the AI within it is monitored, the answer will be more damaging than a missing document. The best-prepared carriers are running tabletop exercises with the four exhibits now, so that the people who own the systems can answer examiner questions directly. They are also identifying which business owners can speak to each system and ensuring those owners have the documentation to back up their answers.

What is not fixed is whether the fall meeting adopts anything and, if it does, which version. The differences that will matter to carriers sit in the exhibit instructions rather than in the headline framework. Until that lands, preparation aimed at the current exhibit structure keeps its value and preparation aimed at guessing the final wording does not.