Insurance Approval Gate NAIC | Moring
The NAIC does not prescribe a technical architecture, but its Model Bulletin direction points insurers toward documented policy, defined accountability, and evidence that oversight is active. An approval gate is the control that most directly produces that evidence in production: every consequential agent action waits for an authorized human decision and leaves a record an examiner can review.
Short answer
The NAIC does not prescribe a technical architecture, but its Model Bulletin direction points insurers toward documented policy, defined accountability, and evidence that oversight is active. An approval gate is the control that most directly produces that evidence in production: every consequential agent action waits for an authorized human decision and leaves a record an examiner can review.
How it works
The NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted in December 2023, is the anchor document for insurance AI governance in the United States, and it deliberately avoids prescribing technology. Program guidance based on the bulletin notes that it does not establish a single technical architecture; instead it points toward a governance structure of documented policy, defined accountability, and evidence that oversight is active rather than aspirational. An approval gate is the control that converts that direction into runtime behavior.
The bulletin's premise is that existing law already applies. It reminds insurers that decisions or actions impacting consumers made or supported by AI systems must comply with all applicable insurance laws and regulations, including unfair trade practices and unfair discrimination. Whatever a carrier automates through agents, the legal standard for the resulting decision does not change, which is the logic behind a gate that keeps a named human in the loop for the decisions where compliance judgments matter most. The bulletin's consumer-protection premise is what elevates operational controls above documentation: a decision affecting a policyholder is judged by the same standard whether a person or a model made it.
Adoption has been broad. Sector summaries report that roughly half of the states have adopted the bulletin, with additional states maintaining insurance-specific AI regulation and guidance, and that non-adopting states still apply bulletin-aligned expectations through market conduct examinations. That breadth means a carrier operating across state lines cannot scope its oversight program to the strictest single jurisdiction and ignore the rest; consistent runtime controls are the practical answer to a patchwork. For multi-state carriers, the practical strategy is to build to the bulletin's governance structure once and apply it uniformly, because jurisdiction-by-jurisdiction tailoring does not scale.
What examiners look for is operation, not documentation. Program guidance notes that examiners tend to look for proof that the policy is operational, meaning a person or committee responsible for it, a record of decisions made under it, and evidence that oversight functions as described. An approval gate produces exactly that: for each consequential claims, underwriting, or pricing action proposed by an agent, a held decision, a named approver, and a retained outcome. The bulletin's expectation of an AI systems program with senior-management accountability, risk controls, and testing maps naturally onto gate policy that defines thresholds, approver roles, and evidence.
Regulator activity is also moving toward examination readiness. Reporting on the NAIC's summer national meeting describes work on an AI Risk Evaluation Supplement that a dozen state insurance departments are piloting to assess insurer AI governance practices, alongside the theme that insurers remain responsible for the technologies, data, and third parties they use. Against that backdrop, runtime controls that produce their own evidence give carriers a materially better examination posture than policy documents alone. Pilots of that kind signal where examination attention is heading, and carriers with runtime controls already in place tend to find the questions answerable.
An approval gate also answers the bulletin's third-party dimension. Where agents act on vendor models or vendor tools, the carrier remains responsible for the decisions they produce, and the gate's record shows the carrier's own control applied on top of the vendor component: the proposed action, the policy version, and the human decision are the insurer's artifacts regardless of where the model came from. Where a vendor model drives the proposal, the gate record still shows the insurer's own control applied, which supports the responsibility the bulletin assigns.
For carriers designing to NAIC expectations, the mapping is direct. Inventory the agents and their actions; define which action classes require human authorization; deploy the gate at the action boundary for those classes; and retain per-decision evidence. That sequence operationalizes the bulletin's governance structure without waiting for any single state's implementation choices, and it produces the demonstration material a market-conduct examination will actually request. The gate also gives internal audit a runtime surface to test, which strengthens the third line of defense the bulletin's program structure presumes.
Frequently asked questions
Does the NAIC require an approval gate specifically?
No, and that is deliberate. The bulletin sets governance expectations rather than prescribing a technical architecture: documented policy, defined accountability, and evidence that oversight is active. An approval gate is simply the control that most directly produces that evidence at runtime, by holding consequential agent actions for a named human decision and retaining the record. Carriers choose the mechanism, but the operational-oversight expectation the bulletin points toward is what the gate satisfies.
How does the bulletin apply if my state has not adopted it?
Adoption varies, but the expectation travels. Reporting on bulletin adoption notes that a large share of states have adopted it, several more are in progress, and non-adopting states still apply bulletin-aligned expectations through market conduct examinations. Since carriers write business across many state departments, building consistent runtime controls such as an approval gate against the bulletin's governance structure is more practical than tailoring oversight to each jurisdiction's current posture.
What exam evidence does an approval gate provide under NAIC expectations?
For each consequential action proposed by an agent, the gate retains the proposed action and arguments, the policy version applied, the approver identity, and the outcome. That is precisely the proof examiners seek that a governance policy is operational: a person responsible, decisions recorded under the policy, and oversight functioning as described. The record also covers the bulletin's third-party dimension, because the insurer's control is shown applied on top of vendor models or tools.
Where does an approval gate fit in an AI systems program?
Within the program structure the bulletin describes, the gate implements the operational controls portion: constraints on automation, monitoring, escalation, and reporting. The written program defines governance, risk management controls, and internal audit across the lifecycle; the gate enforces the automation constraints in claims, underwriting, and pricing workflows and generates the decision records that internal audit and examination readiness both consume. That division keeps the written program and the approval gate producing the evidence NAIC-aligned examination expects from an insurer.