NAIC Panel on GLMs and AI Governance, With an Evaluation Tool Update

The NAIC’s Big Data and Artificial Intelligence Working Group is scheduled to meet on July 22, 2026. Two items are on the agenda: a panel discussion from the actuarial organizations on generalized linear models and AI governance, and a progress report on the AI Systems Evaluation Tool pilot. The meeting is a checkpoint on the way to the NAIC Fall 2026 National Meeting, where an updated version of the tool is scheduled for adoption consideration, and carriers should pay attention to both the pilot feedback and any signals about how the tool will be used in examinations.

The AI Systems Evaluation Tool creates no new law. Its exhibits are optional, and an examiner works through them on top of the market conduct, financial analysis, and financial examination work a state already does. The pilot now covers 12 states, and what it tests is narrow: whether the questions are practical to ask, and whether the documents carriers hand back answer them.

The July meeting matters as a mid-pilot readout, not as the final one. The pilot runs through September, the working group meets again at the Summer National Meeting in August, and the tool is revised and reissued for review in the autumn, before the fall meeting takes the adoption question up in November. Pilot states have been running the questions against real insurers and real systems, and their feedback could change which documents are treated as sufficient, how high-risk systems are defined, and how examiners interpret governance gaps. Every month of that stretch is still a month for closing gaps quietly; if the tool is adopted, the same gaps surface in an exam file instead.

The pilot update is the item that feeds the draft put up for adoption in the fall. Carriers that cannot attend should still review the materials when they are posted, because the pilot comments often reveal which questions regulators found most difficult to answer and which documentation gaps are most common.

Confirm before the meeting that the inventory and governance documentation built for the four exhibits is current and internally consistent. The place to look first is not a missing policy; it is a disconnect between the systems listed in Exhibit A and the governance oversight described in Exhibit B. If the inventory and the committee do not agree, the examiner will notice. Carriers should also check that, for each high-risk system, they can name the data categories it consumes and say of each one whether it is internal or arrives from a specific vendor, because that is the form Exhibit D asks for.

Someone on the team is worth assigning to the call itself, because the panel discussion may offer early signals about how regulators view agentic AI and vendor responsibility. These signals can inform not only NAIC preparation but also responses to state AI legislation and litigation that is already moving through the courts. There is no transcript to catch up on afterward. The agenda and materials go up before the call, and the minutes turn up later as an attachment to a subsequent meeting, which is why the live hour is the one worth a seat.

If you send someone, send whoever owns the model inventory rather than a compliance observer. The signals worth catching concern how regulators talk through agentic systems and vendor responsibility, and they surface as asides in panel discussion rather than in the published materials. Brief that person on the four exhibits beforehand so they can tell which asides are worth writing down.