Boring Insurance Agency
A conference table at dusk with a laptop showing a checklist, a black binder, reading glasses and a printed map, a domed government building through the window

What does the NAIC AI model bulletin mean for an insurtech or an AI vendor selling to insurers?

The bulletin is addressed to insurers, not to software vendors, so it does not regulate you directly. It reaches you through the insurer: it expects the insurer to check out third-party AI before using it, to ask for audit rights and regulator cooperation in your contract, and to be able to hand a regulator that contract and your testing records. As of April 1, 2026 the NAIC map lists 25 jurisdictions as adopted. If you sell models or data to insurers, expect these as questions and clauses, and keep the answers ready.

/ Start here

Tell us the situation.

Already a client and need a certificate, ID card, policy change or to report a claim? Send a service request.

One licensed human replies the same business day — not five agents, not an auto-responder. If the policy you have is already the right one, we will tell you.

We use this to quote and service your insurance, and we do not sell it or pass it to lead networks. Privacy policy.

Kevin Kelsey, Founder | Principal Agent
/ Who handles this

Kevin Kelsey

Founder | Principal Agent

A software developer as well as an agent. Kevin owns and runs TrainRec, a fitness app with a built-in AI coach, and has placed insurance for AI apps, so he has seen how an underwriter reads a model that gives personal guidance, more like a professional than like software. The question every AI app founder brings, which policy answers when the output is wrong, is one he has had to answer for his own product.

/ Don’t want to read all this?

Send us what you have. We’ll tell you if it’s right.

Your current policy paperwork, a renewal notice, or nothing at all. A licensed agent reads it and replies the same business day.

What the bulletin is, and what it is not

The NAIC is the group of state insurance commissioners, and on December 4, 2023 it adopted a model bulletin on the use of AI systems by insurers. A model bulletin is a template. It has no force until a state insurance department issues its own version to the insurers it licenses, so the legal text that matters to a business is the state’s bulletin, not the NAIC’s.

What the template says is short in principle. Decisions that affect consumers must follow the insurance laws on unfair trade practices and unfair discrimination, whether a person or an AI system made them. Insurers are expected to write down how they govern AI, test it and watch it for errors, bias and drift, and to be ready for a regulator to ask for the evidence. It does not ban any technique and it does not prescribe a method. It says outright that an insurer may show compliance by other means.

It also defines AI broadly. An AI system is a machine-based system that produces predictions, recommendations, content or other output that influences decisions, with varying levels of autonomy. A rate-scoring model, a claims triage tool, a chatbot that steers a quote and a fraud flag all fit.

Who it binds: insurers, not you

The bulletin is addressed to insurers that hold certificates of authority in the state. Its program expectations are for the insurer: governance run by senior management accountable to the board, controls matched to how much harm a decision could do to a consumer, and records of data, validation and monitoring. A software vendor, a data provider or a managing general agent is not the addressee.

That does not make it irrelevant to you. The program is supposed to cover AI systems across underwriting, pricing, claims, marketing and fraud detection whether the insurer built them or a third party did. The insurer cannot satisfy that for the system you sold it without something from you.

How it reaches a vendor: the contract and the questionnaire

The third-party section asks the insurer to do due diligence on the vendor and on its data or AI, and, where appropriate and available, to put two things in the contract. The first is audit rights, or the right to receive audit reports from a qualified outside auditor. The second is a duty to cooperate with regulatory inquiries and investigations about the insurer’s use of the product. Then it asks the insurer to actually use those rights.

The examination section tells the insurer what a regulator may ask for, and the list includes the vendor contract itself. The terms the bulletin names are representations, warranties, data security and privacy, data sourcing, intellectual property, confidentiality and cooperation with regulators. It also lists due diligence done on the third party, any audit of the third party’s compliance, and validation, testing and auditing records, including evaluation of model drift.

In practice that becomes a vendor security and model questionnaire, a new audit clause in the renewal and, if the insurer is asked, a request that you produce your testing documentation. Two states show how the same idea is written down. Colorado’s Regulation 10-1-1 keeps the insurer responsible for its vendors but lets the vendor hand the Division the documents directly. New York’s Circular Letter No. 7 says an insurer may not rely solely on a vendor’s claim that its product does not discriminate, though it need not understand the AI’s detailed inner workings.

Where it stands: states that adopted it and states with their own rules

The NAIC keeps a map. As of April 1, 2026 it lists 25 jurisdictions as adopted, which is 24 states plus the District of Columbia, with Nebraska shown as an issued guidance document. California, Colorado, New York and Texas are listed apart because they already had insurance-specific rules or guidance. The NAIC states that the map does not decide whether a state’s action contains every element of the model.

Two practical points follow. A vendor that sells nationally will meet the same few asks in many states, so building one package of documents is cheaper than answering each insurer from scratch. And a state not on the adopted list is not free of the same expectations, since the unfair trade practice and unfair discrimination laws the bulletin points to apply to insurers regardless of whether a bulletin has been issued. Check the map for the states you serve, because it changes.

What it means for your own insurance

A vendor’s contract with an insurer will often ask it to indemnify the insurer if a regulator finds a problem with the vendor’s model. Technology errors and omissions (E&O) insurance may pay what the law would make you owe, and may carve out liability you took on only by contract, so read the indemnity against the policy before signing. Fines and penalties are commonly excluded.

If the product recommends coverage to the public, you may be an agent as well as a vendor, and the E&O that fits an agent is a different policy. Our insurtech page walks through which of the two applies to which kind of company.

What the regulator may ask the insurer, and what the vendor should have ready

The left column follows the bulletin’s list of what an insurer should expect to be asked about third-party AI. The right column is our reading of what that means for you, not a requirement.

  • Due diligence done on the third party and its data, models or AI systems

    What a vendor should be able to hand over
    A plain description of the model, its intended use, its data sources and its limits
  • Contracts with AI, model and data vendors, including terms on representations, warranties and data sourcing

    What a vendor should be able to hand over
    Your standard terms, with the warranties you will actually stand behind
  • Audits or confirmation of the vendor’s compliance with contract and regulatory duties

    What a vendor should be able to hand over
    An audit report or a right-to-audit process you can live with
  • Validation, testing and auditing records, including evaluation of model drift

    What a vendor should be able to hand over
    Test results, bias and drift checks, and how often you rerun them
  • Data used to build the model: source, provenance, lineage, quality and bias analysis

    What a vendor should be able to hand over
    A data lineage summary and a note on what you cannot disclose
  • Terms on cooperation with regulators

    What a vendor should be able to hand over
    A named contact and a response time for a regulator request routed through the insurer

The bulletin says its goal is not to prescribe specific documentation. Treat the list as the questions you will meet, and answer in the form the insurer’s state accepts.

The detail

What this means for you

  • Treat the bulletin as a contract clause you will receive. The insurer is told to seek audit rights and regulator cooperation from vendors where appropriate. Decide now what audit you will allow, how fast you will answer a regulator’s request, and what you will not disclose.
  • Write one model file. A description of what the model does and does not decide, its data sources, its limits, how it was tested for errors and bias, and how it is monitored for drift. The same file answers Colorado, New York, an adopting state’s bulletin and a customer questionnaire.
  • Say what a person decides. The bulletin ties the strength of controls to how much humans are involved in the final decision. A tool that recommends and leaves the decision to a licensed person sits differently from one that decides alone.
  • Read every indemnity against your policy. A promise to cover an insurer’s regulatory cost may not be insured. Send the clause to your agent before you sign it.
  • Check which states your customers sit in. The map lists adopted states and the four with their own rules, and it changes. Do not assume a state is covered or uncovered from this page.
/ Start here

Want this handled? Get a quote.

One licensed human replies the same business day. We do not sell your details or pass them to lead networks. Privacy policy.

Related questions

Does the NAIC bulletin apply to me if I only sell software to insurers?

Not directly, since it is addressed to insurers. It reaches you through the insurer, which is expected to review third-party AI, to seek audit rights and regulator cooperation in your contract, and to be ready to show a regulator that contract. Expect those as questions and clauses from your insurer customers.

Is the bulletin a law?

No. It is a model that the NAIC adopted on December 4, 2023. It takes effect in a state only when that state’s insurance department issues its own version, and the NAIC says its map does not decide whether a state’s version contains every element of the model.

How many states have adopted it?

The NAIC’s map, status as of April 1, 2026, lists 25 jurisdictions as adopted: 24 states and the District of Columbia, with Nebraska shown as an issued guidance document. California, Colorado, New York and Texas are listed separately with their own insurance rules. Check the current map, because it changes.

What does Colorado require of vendors?

Colorado’s Regulation 10-1-1, effective November 14, 2023 for life insurers, keeps the insurer responsible for meeting every requirement when it uses third-party vendors and requires a documented process for choosing and overseeing them. It lets the vendor give the requested documents directly to the Division on the insurer’s behalf. The NAIC map shows later amendments effective October 15, 2025.

What did New York say about AI vendors?

Insurance Circular Letter No. 7 of July 11, 2024 says insurers keep responsibility for AI and outside data used in underwriting and pricing, may not rely solely on a vendor’s claim of non-discrimination, and should have vendor standards and, where appropriate, audit rights and cooperation terms in contracts. They need not understand the AI’s detailed inner workings.

Does the bulletin ban any kind of AI or require bias testing?

It bans no technique and prescribes no method. It encourages verification and testing to find errors and bias, expects controls proportionate to the risk, and says an insurer may show compliance through other means. The unfair discrimination laws it points to are what carry the force.

Does my insurance cover a regulator’s finding about my model?

Often only in part. Technology E&O may pay what the law would make you owe, may carve out liability assumed only by contract, and commonly excludes fines and penalties. Whether the cost of answering a regulator is covered before a lawsuit exists depends on the wording. Send us the policy and the indemnity clause.

Insurance for insurtechs
Does contingent bodily injury matter here?

Not usually, since an insurance model loses money and does not hurt bodies. It is the coverage for an injury someone says came from relying on an app’s output, and it matters if the product also gives health or safety guidance. The guide explains how it works.

Bodily injury against contingent bodily injury

Need this handled?

Reading about it and having it handled are different jobs, and the second one is ours. We place the accounts other brokers decline — claims history, a cancellation, an unusual operation, a business nobody wants to underwrite twice — and we will tell you when the policy you already have is the right one. Tell us the situation and a licensed human replies the same business day.

/ Start here

Want this handled? Get a quote.

One licensed human replies the same business day. We do not sell your details or pass them to lead networks. Privacy policy.

Written by Kevin Kelsey, founder and principal agent, CA License #6011822, at Boring Insurance. Last updated 2026-11-01. See all guides.

Call/Text(626) 344-2158Quote