Insurance Agent Registry NAIC Bulletin | Moring
The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted December 4, 2023, is principles-based guidance for overseeing AI in underwriting, rating, claims, fraud and marketing. It expects a written AI program, proportionate governance, testing for bias and errors, and third-party accountability. For agents, the registry is the operational layer: each entry carries the owner, approval and evidence the bulletin's program implies.
Short answer
The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted December 4, 2023, is principles-based guidance for overseeing AI in underwriting, rating, claims, fraud and marketing. It expects a written AI program, proportionate governance, testing for bias and errors, and third-party accountability. For agents, the registry is the operational layer: each entry carries the owner, approval and evidence the bulletin's program implies.
How it works
The Model Bulletin on the Use of Artificial Intelligence Systems by Insurers is the anchor document for United States insurance AI governance. Adopted by the association on December 4, 2023, it is principles-based guidance rather than a model law, and it is not self-executing: authority rests with each state insurance regulator that adopts it, and adoption timing and language vary. What the bulletin does is restate, in the supervisor's own words, that existing prohibitions on unfair trade practices, unfair discrimination and unfair claims settlement apply in full when a decision is made or supported by an AI system, and then describe what managing that exposure looks like.
Its architecture is a program, not a technique. The bulletin describes an AI Systems Program with senior-management and board accountability, risk controls proportionate to the use case, testing for errors, bias and unfair discrimination, and documentation an examiner can review, extending across underwriting, rating, claims, fraud and marketing. Guidance materials translating the bulletin emphasize the same implementation lesson: a policy document by itself is rarely sufficient, because examiners look for proof the policy is operational, meaning a responsible person or committee, a record of decisions made under it, and controls that trace back to specific production use cases.
Agents raise the bulletin's stakes because they move the object of governance from output to action. A rating model produces a recommendation a human acts on; a claims triage agent routes the claim, drafts correspondence and, in some configurations, determines coverage position directly. The bulletin's expectations translate into registry practice naturally: each agent entry records its purpose, its accountable owner, the approval that authorized its current build, its permission boundary over policy and claims data, and the decision records it produces, giving the program its use-case-level traceability.
Third-party accountability is the bulletin's most underestimated element. It makes the insurer accountable for AI sourced from a third party, which means vendor-supplied agents embedded in claims platforms, underwriting workbenches or servicing suites enter the same governed record as internal builds, with vendor lineage captured. Governance guides also stress that adoption varies by state, so carriers should confirm current requirements against their own jurisdiction's published guidance rather than treating any single interpretation as universal; a multi-state carrier effectively builds to the strictest adopted posture it faces.
Adoption breadth has grown steadily since adoption began: industry tracking counts twenty-five states plus the District of Columbia on the bulletin, twenty-six jurisdictions including states with their own insurance AI rules, and at least twenty-eight jurisdictions enacting the related Insurance Data Security Model Law that now pairs with AI governance obligations. The association's public topic material reinforces the operational context, describing AI in claims processing, chatbots serving policyholders and predictive approaches to risk, all paths where agents now operate and where examination questions concentrate.
The bulletin's practical translation for an agent registry is a checklist per entry: named owner, committee or named-approver sign-off recorded, risk tier assigned by customer impact, tested controls over data access and tool use, bias and error testing evidence, third-party lineage where applicable, and decision records retained per policy. A carrier whose registry carries those fields for every claims, underwriting and servicing agent has, in effect, implemented the bulletin's program at the use-case level, and can demonstrate active rather than aspirational oversight when the examination request arrives.
Frequently asked questions
What is the NAIC Model Bulletin and when was it adopted?
It is the NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, principles-based guidance adopted on December 4, 2023. It addresses AI across underwriting, rating, claims, fraud and marketing through a documented AI systems program, and it is not a model law or self-executing rule. Each jurisdiction adopts it on its own timing and language, so carriers confirm the posture where they write insurance business while building governance to the demanding end of adoption.
What does the bulletin require for AI systems like agents?
It describes an AI Systems Program with senior management and board accountability, risk controls proportionate to the use case, testing for errors, bias and unfair discrimination, documentation an examiner can review, and oversight of third-party AI for which the insurer remains responsible. For autonomous agents, that translates into registration with an owner, an approved build, a permission boundary over claims and policy data, and retained decision records, so the program is demonstrably operational rather than a policy document.
How does the bulletin treat third-party and vendor AI?
It makes the insurer accountable for AI sourced from a third party, which is the point many carriers underestimate. Vendor-supplied agents embedded in claims platforms or underwriting tools therefore need the same registration, owner assignment, permission boundaries and evidence trails as internally built agents, with vendor lineage recorded. When an examiner questions a vendor agent's decision, the carrier is the accountable party and must produce the record, which is why vendor agents belong in the registry.
How widely has the bulletin been adopted?
Industry tracking places adoption at twenty-five states plus the District of Columbia, with other counts citing twenty-six jurisdictions once states with their own insurance AI rules are included, and at least twenty-eight jurisdictions enacting the related Insurance Data Security Model Law. For a multi-state carrier this makes bulletin-aligned governance functionally national. The registry, with per-agent owners, approvals and evidence, is the structure that satisfies the expectation across jurisdictions without rebuilding it state by state.