Software Probably Set the Price on Your Totaled Car. You Have Until Tuesday to Tell Regulators What to Ask It.

If an insurer has ever handed you a number for a wrecked car and could not explain how it got there, there is a decent chance nobody at that company could either. A model produced it.

State regulators have spent the last six months pulling those models apart. The rulebook they built is open for public comment until close of business Tuesday.

Here is the part worth your attention. When the NAIC last surveyed private passenger auto insurers, it got 193 responses. Of those, 135 were already using AI or machine learning in claims. Fifty-five used it to evaluate loss images. Fifty used it to determine settlement amounts.

Nine used it for claims approvals.

Zero used it for claims denials.

Read those last two lines again. The industry did not build machines to tell you no. It built them to decide how much. A denial is visible, appealable, and a regulatory problem. A number that is nine hundred dollars light is none of those things. It is just the offer.

That survey is from 2023, and its age is the whole point: three years on, nobody has a current, standardized picture of what these systems do. That is what the National Association of Insurance Commissioners is now fixing. Its Big Data and Artificial Intelligence Working Group ran an AI Systems Evaluation Tool through a pilot in twelve states from March to September of this year: California, Colorado, Connecticut, Florida, Iowa, Louisiana, Maryland, Pennsylvania, Rhode Island, Vermont, Virginia and Wisconsin.

The tool is four exhibits. How much AI the insurer uses. How it governs those systems. Details on anything it classifies as high risk, meaning systems that could cause serious consumer or financial harm. And the data feeding all of it. Autobody News, reporting on the pilot in March, said the tool reaches the systems that determine total-loss thresholds, assess damage, and set parts valuations.

Translation: the software that decides whether your car is repairable, and what it was worth if it isn’t.

The working group is taking comments on version 5.0 of the AI Risk Evaluation Supplement until close of business Tuesday, September 29. Its page names the two staffers to send them to. There is a public Webex on October 8 at 11:00 a.m. Eastern. The finished tool is expected to go up for adoption at the NAIC fall national meeting in November.

Smart. Late, but smart. A regulator that can only read the outcome of a claim is reading the last page of the book.

If you have something to say, say it by Tuesday. The most useful comment you can write is not an opinion about AI. It is a specific question you think an examiner should be able to force an insurer to answer: which comparable vehicles the model chose, what condition adjustment it applied and why, and how often its number lands below what the same car sells for locally.

If you have a claim open right now, none of this rescues it. Do the thing that does. Ask for the full valuation report, not the summary. Ask for the list of comparable vehicles and the adjustment applied to each. Then check whether your policy still carries an appraisal clause, because one driver used hers to collect more than $4,000 after her class action went nowhere.

And if your carrier’s opening number tells you what it thinks of you, re-shop the policy once the claim closes. Our auto insurance estimator gives you a sane starting figure, and the auto insurance hub has the coverage mechanics.

One caveat on geography. Your state not being in the twelve does not mean your insurer skipped this. Carriers write across state lines, the tool is built for nationwide adoption in November, and the models being examined are the same ones running your file.

How Candid Yak makes money. Some of the products we write about pay us if you apply or sign up through our links. That never changes our verdict, our rankings, or the numbers in this article. We call a bad deal a bad deal whether it pays us or not. Some brands shown in our comparison tools are placeholder examples while we finalize partner agreements, and we label them as such.