Book a demo
Topics · GOVERNANCE-AGENTS

Insurance Approval Gate NAIC Bulletin | Moring

The NAIC Model Bulletin defines a broad AI systems program, covering governance, risk controls, testing, and third-party oversight across underwriting, rating, claims, fraud, and marketing. An approval gate operationalizes its hardest requirement, constraints on automation, by holding consequential agent actions for named human decisions and producing the runtime evidence the bulletin's examination expectations anticipate.

On this page
  1. Short answer
  2. How it works
  3. FAQs
  4. Related reading
  5. Sources

Short answer

The NAIC Model Bulletin defines a broad AI systems program, covering governance, risk controls, testing, and third-party oversight across underwriting, rating, claims, fraud, and marketing. An approval gate operationalizes its hardest requirement, constraints on automation, by holding consequential agent actions for named human decisions and producing the runtime evidence the bulletin's examination expectations anticipate.

How it works

The Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted by the NAIC on December 4, 2023, defines the terms of engagement for carrier AI programs. It is not a standalone statute; it states explicitly that it does not establish new requirements, instead reminding insurers that existing laws, including unfair trade practices statutes and rating and underwriting requirements, apply fully to decisions made using AI. Its definition of an Artificial Intelligence System is intentionally wide, capturing machine learning models, generative tools, and other predictive algorithms across underwriting, rating, claims, marketing, and fraud detection. The breadth of that definition matters for gate design, because it puts ordinary predictive models used in rating and claims inside the same program expectations as newer generative tools.

The bulletin describes an AI Systems Program with specific elements: senior-management and board accountability, risk management controls, testing for errors, bias, and unfair discrimination, and third-party oversight where the insurer remains responsible. Program guidance for chief risk officers observes that some expectations need written governance only, while others, such as constraints on automation, monitoring, escalation, and reporting, also need operational controls that produce evidence. That split is exactly where an approval gate sits.

Constraints on automation are the bulletin's most runtime-oriented expectation, and an approval gate is their direct implementation. The control interrupts an agent immediately before it executes a designated high-risk action and holds execution until a human with the appropriate authority explicitly decides on the proposed action, evaluating the resolved parameters and context against policy before any side effect occurs. For a carrier, that means a settlement above the delegated band, a coverage denial, or a bind decision waits for its authorized reviewer, and the hold is technical rather than procedural. Holding the action is what distinguishes a constraint from a monitoring alert, which fires after the fact and leaves the outcome in place.

Evidence expectations are equally concrete. Examination-oriented guidance asks a pointed question of insurers: pick one AI interaction or tool call and show what the system was permitted to do, what it did, and which policy version applied, and confirm that the escalation path produces a retained record of the human decision. An approval gate answers that by construction, because every gated decision is recorded with its action, arguments, policy version, approver, and outcome, in a form a state department can review.

Adoption dynamics shape how carriers should treat the bulletin. As of mid-2026 summaries, two dozen states and more have formally adopted it, with additional states in progress, and non-adopting states still applying bulletin-aligned expectations through market conduct examinations. Legal commentary adds that the bulletin continues to evolve and that insurers should expect AI governance and its supporting cybersecurity controls to remain a regulatory priority, with proactive alignment reducing future compliance and examination risk. Record completeness is what makes the distinction verifiable rather than asserted, so evidence wiring belongs in the gate's core design.

The bulletin's testing expectations also connect to the gate. Testing for errors, bias, and unfair discrimination is required before and during use, and the gate contributes a runtime complement: by classifying decisions by consequence and routing the high-stakes ones to licensed human review, the carrier reduces the population of ungated automated decisions where a testing gap would translate directly into consumer harm. Adverse determinations, in particular, are the class where sector runtime designs require licensed human review before the agent responds at all. Carriers that align early avoid the position of demonstrating governance for the first time in response to a specific examination question.

For a carrier mapping its program to the bulletin, the approval gate is the artifact that binds the written program to production behavior. The AIS Program document says which decisions require human authorization; the gate makes that true on every call; and the retained records demonstrate both, which is the combination state examiners are increasingly equipped to verify.

Frequently asked questions

What does the bulletin mean by constraints on automation?

It refers to operational limits on what AI systems may do without human involvement, one of the expectations that requires controls producing evidence rather than written policy alone. An approval gate implements this by holding designated consequential actions, such as settlements above authority bands or coverage denials, until an authorized human decides, with the resolved parameters evaluated against policy before execution. The resulting hold-and-decide records are the evidence the constraint exists.

Is the bulletin binding on insurers?

Not directly. The bulletin is guidance that becomes enforceable when individual states adopt it through their own regulatory channels; a large share of states have done so, with others in progress, and non-adopting states still apply bulletin-aligned expectations in market conduct examinations. The bulletin itself reminds insurers that existing insurance laws apply fully to AI-supported decisions, so the underlying obligations exist regardless of adoption status.

How does the gate satisfy the bulletin's evidence expectations?

Examination guidance asks insurers to show, for a chosen AI interaction or tool call, what the system was permitted to do, what it did, and which policy version applied, plus a retained record of the human decision on escalation. An approval gate produces all of that for every gated decision: action, arguments, policy version, approver, and outcome. The evidence is generated by the control itself rather than assembled retrospectively for an exam.

How does the bulletin's testing requirement relate to the gate?

The bulletin expects testing for errors, bias, and unfair discrimination across the AI system lifecycle. The gate complements testing at runtime by routing the highest-consequence decisions, notably adverse determinations, through licensed human review before the agent responds, so a testing gap cannot translate directly into consumer harm on an automated path. Override and approval data from the gate also feeds back into the testing program as production evidence. The pairing keeps the bulletin's testing intent enforced in production through the carrier's approval gate.

Sources

  1. content.naic.org
  2. vijil.ai
  3. openlayer.com
  4. trussed.ai
  5. thedataexperts.us
  6. plantemoran.com
  7. thedataexperts.us
  8. trussed.ai
  9. - artificialintelligenceact.eu
  10. - eur-lex.europa.eu