The AI exam supplement now has an official record

The Big Data and Artificial Intelligence (H) Working Group’s summary of its August 13 session is now posted, and it runs to a single page. On that page the supplement’s new name enters the official record, the pilot’s reach turns out to be wider than a market conduct file, and the two steps we have been putting on your calendar for later in the year go unmentioned.

This week’s story: the record settles the name and goes quiet on the calendar

The summary reads as nested minutes, so indentation rather than line order decides which meeting used which name. “AI Risk Evaluation Supplement (formerly known as the AI Systems Evaluation Tool)” occurs exactly once, in the July 22 entry. The August 13 entry uses the new name with nothing attached, and the record gives the change no date. Whether the instrument’s text moved with the name is a separate question, unanswered in the summary; our standing account of what each exhibit asks holds that side.

Pilot states, the August 13 entry says, are using the supplement “in support of a mix of market conduct exams, financial exams, and financial analyses, as well as in the context of a more general regulatory inquiry.” We read four settings there, and the arithmetic is ours: the NAIC printed the list without a count. A general regulatory inquiry, on our reading, arrives with no examination attached. If your procedure sends AI requests to whoever handles market conduct, it has three gaps.

Adopting the July 22 minutes was the only action on August 13, and for the supplement the record shows a pilot update with neither an adoption nor an exposure attached. A revision of the instrument across September and October and adoption consideration in November, both carried here since July, appear nowhere in it. They rest on one NAIC document, the pilot project summary, and nothing published on August 13 renews them.

Three more items rest on less. The instrument’s next version number, a comment window inside September, and a survey deadline for pilot companies reached this newsletter in July from one signed Coverager column by the head of an AI claims vendor. Issue 03 attributed them at the time, and the August 13 record corroborates none of the three, so they leave this issue’s calendar. Our ownership matrix for insurance AI governance puts a named function against every artifact a department can ask for.

What to do this week

• Ask whoever owns exam readiness which of the four settings would reach your book first, and get the answer in writing.
• Check whether your regulatory response procedure has a route for a request that arrives with no examination attached. The summary’s fourth setting is a general regulatory inquiry, and there is no exam number to file it under.
• Open your team’s AI regulatory calendar and write the source beside every entry for this instrument. The September and October revision and the November adoption item come from the NAIC’s pilot project summary; a comment window inside September comes from a vendor executive’s column, and the August 13 record did not supply one.

On the Docket

• September 2026: The last month in which the twelve pilot states use the supplement. The August 13 record and the NAIC’s pilot project summary agree on the window: the pilot opened in March, twelve states are taking part, and it runs through September. Carriers writing in those twelve, plus anyone who wants the current text in hand before the revision, which rests on the pilot project summary alone. NAIC meeting summary

• November 2026: The Fall National Meeting is where the revised supplement would be taken up for adoption. One document carries this, the NAIC’s pilot project summary; the August 13 record does not mention it. Everyone the pilot has not reached yet. NAIC pilot summary

• December 1, 2026: Colorado life insurers file their next annual report under Regulation 10-1-1, and this is the first one the Division’s quantitative testing waiver does not reach. Revised Bulletin B-10.004, adopted October 22, 2025, confines that waiver to the reports due December 1, 2024 and December 1, 2025, and says subsequent ones “will be expected to include a description of the quantitative testing conducted.” The reason the bulletin gave was that the Division had not yet adopted the regulation establishing that testing, and that was still the position at our last check on August 2. Colorado life writers. Auto and health writers as well, though the waiver never covered them and Section 5.A.11 was already in force for their first annual report on July 1, 2026. Colorado DOI adoption notice

This week in brief

Governance

AM Best told the working group on August 13 that governance has to scale with each use case’s risk as insurers move up from predictive models to generative and agentic ones, and that doing so takes validation, oversight by a person, monitoring, documentation, authorization limits, and a way to roll a system back. That is a rating agency’s expectation given to the working group rather than an NAIC position, and its value is the prospect of a ratings analyst asking about the systems an examiner already inspects. NAIC meeting summary

One data point

Four settings, by our count. Only one of them is a market conduct exam, and market conduct is where readiness work, our own coverage of it included, has concentrated. We cannot tell from this summary what the other three would ask for. Our evidence pack for reconstructing a single AI-assisted decision is narrower than any of the four: it holds one transaction, with the source facts, the model version, the human action and the outcome kept connected, and its own scope note says plainly that this is a transaction record and no substitute for an organization-wide AI inventory.

The public record will not show which of the four tracks the supplement actually came through in any given state. Carriers inside the twelve pilot states may already know. That answer can only come from a reader, and the way to send it is to reply to the email this issue arrived in.

– The Editor, InsureAI Wire