NAIC's AI Risk Evaluation Supplement Faces Two Adoption Votes in Dallas

NAIC By InsureAI Wire. Published November 11, 2026.

Source: NAIC official document, content.naic.org (opens in a new tab)

On November 16 in Dallas, the NAIC’s Big Data and Artificial Intelligence (H) Working Group meets to decide whether to adopt version 7.0 of the AI Risk Evaluation Supplement. The next day its parent, the Innovation, Cybersecurity, and Technology (H) Committee, has an adoption item on the supplement. The committee’s November 10 agenda draft lists the item; its October 29 draft did not. The supplement holds the exhibits regulators draw on when they question an insurer about its AI systems, and every page of 7.0 is still marked DRAFT.

The working group’s page posts 7.0 as two PDFs, an edited copy with notes in its margins and a clean copy. The page gives no posting date. It had no version 7.0 on November 3, as our report on the October 8 minutes noted, so the files went up some time after that.

The group’s meeting agenda, dated November 4 and listed on the page as updated November 5, gives the session an hour from noon on November 16 at the Dallas convention center. Item 4 reads “Consider Adoption of the AI Risk Evaluation Supplement v7.0.” Three items come before it. The group first takes up the minutes of its November 2 meeting. Coral Manning of Wisconsin then gives a summary report on the supplement’s pilot process, and interested parties sum up the written comments the group received.

The committee’s agenda, in a third draft dated November 10, runs from 12:15 to 1:30 p.m. on November 17. Under item 2 the committee decides whether to adopt its working groups’ reports, this group’s among them. Adopting a group’s report is a separate act from adopting the supplement. Item 3, listed under the committee’s chair, Commissioner Michael Yaworsky of Florida, reads “Consider Adoption of the AI Risk Evaluation Supplement.” Neither the November 10 draft nor the October 29 draft gives a reason for the difference. Item 3 names no version. We read it as 7.0, because that is the text the working group considers the day before.

Set against version 6.0, three changes in the new text bear most on what an insurer could be asked. Version 6.0 already let regulators exclude some or all GLMs “if appropriate,” citing the oversight that ratemaking already provides. That exclusion reached only inquiries under Exhibits B through D, which cover an insurer’s AI program, its high-risk AI models and the data those models use. Version 7.0 keeps the basis, the “if appropriate” and the regulator’s discretion. What changes is the reach. Regulators may now exclude some or all GLMs, and some or all other predictive models, from the supplement as a whole. Rule-based systems, deterministic algorithms and fixed mathematical formulas are the text’s examples of the latter.

A new section, Human in the Loop, gives regulators three points they “should consider” when they assess human oversight controls. The wording is guidance to regulators, not a requirement on insurers. One point is authority: whether the reviewer can reverse, modify or escalate an AI-assisted decision, and whether the record shows that overrides actually occur and are documented with reasons. Another is whether the reviewer has the experience or qualifications to evaluate how the AI performs. The third asks whether the review is designed to support effective oversight, and whether that can be shown through measures such as override and escalation rates.

Exhibit C also loses item 10a. Where applicable, the item had asked whether a company evaluates a model’s inputs, its derived variables or its outputs for possible proxy relationships or effects that could feed unfair discrimination or harm consumers. A margin note in the edited copy reads “Deleted to address APCIA, BCBSA, and NAMIC’s input.”

The committee’s agenda does not say what becomes of its item 3 if the working group does not adopt 7.0 on November 16.

Status as of November 10, 2026. Neither vote has taken place. Both agendas are drafts, and the committee’s is already in its third version.

Share

Go deeper

Keep tracking

Information aggregation and analysis, not legal advice. See our disclaimer.