What insurers asked for in the last round on the NAIC AI supplement

The NAIC closed comments on version 6.0 of its AI Risk Evaluation Supplement on October 30, and the 20 letters received are now posted as one file. Slides for the Big Data and Artificial Intelligence (H) Working Group’s November 2 Webex set out a timeline for this month. It runs from revising the supplement based on the comments to “Consider v. 7.0 for adoption at Fall NM,” the NAIC’s national meeting in Dallas. No version 7.0 has been posted.

This week’s story: what the 6.0 letters ask the working group to change

Six of the 20 are industry letters. On GLMs, APCIA wants them out of Exhibits B–D, where 6.0 leaves that “discretionary” (p.70). The Insurtech Coalition would exclude GLMs behind filed rates and risk classifications “throughout the Supplement, including Exhibit A” (p.34). ACLI would have regulators exclude GLMs and other rule-based, deterministic or transparent models “from the Supplement” (p.38). NAMIC wants GLMs out of scope or, failing that, ratemaking GLMs “explicitly out of scope” (p.78).

ACLI and NAMIC would delete item 10a, the proxy question; APCIA would tie it “to applicable law.”

On Adverse Consumer Impact, NAMIC would restore “in a manner that violates those standards,” the words 6.0 cut, lest legal cancellations, non-renewals and premium increases count (p.79). BCBSA, between 6.0 and NAMIC, would cover adverse outputs inconsistent with “applicable regulatory standards or with the established criteria the company applies” (p.123), citing the model bulletin our state map tracks.

What to do this week

Our suggestions, for the weeks before 7.0 is posted:

• If you made Issue 15’s list of GLMs and the rate filings that use them, add a column for each model’s use: ratemaking, risk classification, other underwriting, claims or other. Some of the GLM proposals above turn on what a model is used for, and the column shows which models each would leave in scope.

• Write down the criteria each AI use applies to its decisions. Under BCBSA’s wording, an adverse output that breaks your own established criteria would be an Adverse Consumer Impact even where it breaks no regulatory standard. A written set shows ahead of time which outputs that test would catch, and under any wording it records which adverse outcomes followed your own rules.

On the Docket

• November 5, 2026: The plaintiffs’ class certification motion is due in Barrows v. Humana. Their request to add naviHealth as a defendant was still undecided when Georgetown’s litigation tracker last updated the case on October 21. October 15 scheduling order · Case tracker

• November 16, 2026: The working group meets in Dallas from noon to 1:00 PM CT, with no agenda posted. The Innovation, Cybersecurity, and Technology (H) Committee, to which it reports, meets November 17 from 12:15 to 1:30 PM CT, also without a posted agenda. Fall National Meeting agenda

• November 16, 2026: Earlier the same day, at 8:00 AM CT, the NAIC’s Third-Party Data and Models (H) Working Group is set to take up item 2 of its draft agenda: “Discuss and Expose the Property/Casualty (P/C) Pricing and Underwriting Framework and Confidentiality Pathway.” Draft agenda

• November 21, 2026: At the annual meeting of the National Council of Insurance Legislators (NCOIL), the NCOIL-NAIC dialogue lists an “Update on Work of NAIC’s Big Data and Artificial Intelligence (H) Working Group” (schedule as of October 27). NCOIL schedule

This week in brief

Business Lines

CMS sent letters between January and March 2026 to four contractors that run prior authorization in its WISeR model for Original Medicare, a file the Electronic Frontier Foundation (EFF) posted in its FOIA suit shows: a corrective action plan demand and a Performance Notice of Non-Compliance to Innovaccer, and requests for information, which find no breach, to Genzeon, Virtix Health and Zyter. None of the letters shows CMS holding back payment, suspending a contractor or ending a contract; our October 28 report goes through them.

EFF describes a new batch of WISeR records from the same suit in its post New Records Obtained by EFF Expose More Problems with Medicare’s AI Experiment. The suit also produced the records behind the MedCity News story 7 Potential Problems with Medicare’s AI Prior Authorization Experiment.

eHealth, an online health insurance marketplace, surveyed more than 1,000 U.S. adults through SurveyMonkey from October 7 to 9. Medical Economics reports the result: 82% of Americans back a law requiring human review of AI claim denials, survey finds.

More outside reading is on our Elsewhere page.

From the Guides

For auto insurers. Our October 28 guide traces a driving record through rating, underwriting and claims and sets out the rules that attach at each step: AI in Auto Insurance: Follow the Driving Data From the Car to the Premium.

If you have read the 6.0 letters, reply with the request you expect 7.0 to take up.

– The Editor, InsureAI Wire