Does the bulletin ban AI in claims?

No. It does not prohibit AI. It moves AI use from 'allowed' to 'examined' — carriers may use AI but must govern it and be able to demonstrate how each decision was made.

How does a carrier become exam-ready for AI in claims?

By keeping an AI governance program and, for each AI-assisted decision, a documented basis: the reasoning, the sources and policy language cited, bias monitoring, and a record of human review. That documented basis is what an examiner inspects.

Resource · Regulation

The NAIC AI Model Bulletin, explained

Q: What is the NAIC AI Model Bulletin?

It is guidance from the National Association of Insurance Commissioners setting out how insurers are expected to govern the artificial intelligence they use in decisions, including claims. States that adopt it expect carriers to have an AI governance program and to be able to show it during a market-conduct examination.

Home › Trust & evidence › NAIC AI Model Bulletin

A plain-language guide for claims and compliance leaders: what the bulletin asks of carriers using AI in claims, where it applies, and what it takes to be exam-ready.

What it is

The NAIC AI Model Bulletin is guidance from the National Association of Insurance Commissioners that tells insurers how they're expected to govern the artificial intelligence they use in decisions — including claims decisions. It has been adopted by 24 states, plus the District of Columbia, and that number continues to grow.

What changed: from "allowed" to "examined"

The bulletin does not ban AI in claims. What it changes is the posture: using AI is allowed, but a carrier is now expected to have an AI governance program and to be able to show how each AI-assisted decision was made if a regulator asks. In practice, a market-conduct examiner can inspect your AI governance the same way they inspect claims handling. AI moved from something you could quietly adopt to something you may be examined on.

What it expects of a carrier

A written AI governance program — roles, oversight, and controls for how AI is used.
Documentation of the AI systems in use and what they do.
Testing and monitoring, including for unfair discrimination / bias.
A remediation path — the ability to inspect, correct, override, and document when something goes wrong.
The same expectations extend to third-party AI vendors a carrier relies on.

What "exam-ready" actually looks like

The thing an examiner ultimately wants is a defensible record per decision. For each AI-assisted claims decision, that means being able to produce: the reasoning, the policy language and sources it relied on, evidence that bias was monitored, and a record that a qualified person reviewed it. If you can hand that over, you're in a strong position; if your AI tool only produces a decision and not the basis for it, you're exposed.

This is the gap Velerian was built to close. Our governed claims agents produce the decision and its documented basis together — a policy citation and full evidence trail with every coverage interpretation, denial, fair-claims-handling review, and fraud screening — so the output is built for an exam, not just a demo. See how governed AI agents work or read the white paper.

Frequently asked

Does the bulletin apply to me if my AI is from a vendor?: Generally yes — the governance expectations extend to AI provided by third parties. Carriers are expected to oversee vendor AI, not treat it as a black box.
Is this the same as a law?: It is guidance a state's insurance department adopts and applies; some states pair it with their own statutes or circulars (for example, New York's Circular Letter No. 7 and Colorado's SB 21-169). The practical effect in adopting states is examination scrutiny on AI use.
What's the fastest way to be ready?: Make sure every AI-assisted decision carries its evidence trail — reasoning, citations, bias monitoring, and human sign-off — and keep an AI governance record. That documented basis is what an examiner inspects.

← Back to Trust & Evidence · Glossary of terms