Book a demo
Topics · GOVERNANCE-AGENTS

Insurance Audit Trail NAIC | Moring

The NAIC's framework for insurer AI use makes the audit trail central: carriers must maintain a written governance program, evaluate AI systems by risk, and retain documentation for regulatory review, which means proving controls operated rather than describing them. Agent activity in claims, underwriting, and customer communications sits squarely in scope where a state has adopted the model bulletin.

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

Short answer

The NAIC's framework for insurer AI use makes the audit trail central: carriers must maintain a written governance program, evaluate AI systems by risk, and retain documentation for regulatory review, which means proving controls operated rather than describing them. Agent activity in claims, underwriting, and customer communications sits squarely in scope where a state has adopted the model bulletin.

How it works

The National Association of Insurance Commissioners has shaped insurer AI expectations through its model bulletin on the use of artificial intelligence systems by insurers, adopted in December 2023 and taken up state by state since. The framework is deliberately short and structural: it asks insurers to maintain a written governance program, to evaluate AI systems in proportion to their potential consumer impact, to define named accountability for systems and outcomes, and to retain documentation that the program operated. The fourth expectation is the one carriers underinvest in, because the bulletin does not ask for a document describing controls, it asks for evidence the controls ran.

Adoption is the operative variable. There is no single national NAIC examination: where a state has adopted the bulletin, its supervisory examination is where expectations are tested, and the adopting state's version governs. More than half of US states plus the District of Columbia have adopted the bulletin, while several large states run their own insurance-specific AI frameworks built on the same core expectation. A carrier's audit trail design should therefore assume examination rather than hope for absence, and build records that satisfy the strictest adopting version it operates under.

Scope reaches agentic systems directly. Agents acting in claims, underwriting, and customer communications fall within what the model describes wherever the state has adopted it, and they are usually the highest-impact systems in the carrier's inventory. The bulletin's risk-based structure expects heightened governance for uses with greater potential to affect consumers, which maps naturally onto the audit trail: claim decisions and coverage determinations need action-level records with named approvals, while lower-impact uses may carry lighter but still demonstrable records.

Examination infrastructure is being built now. The NAIC began piloting its AI Systems Evaluation Tool, since renamed the AI Risk Evaluation Supplement, with twelve participating states, and the central evidence question in any such review is reconstructability: can the insurer show one decision, a claim or a coverage determination, end to end, with approvals and records intact. Adoption surveys reinforce the trajectory; most responding insurers report using, planning, or exploring AI, and a substantial share of models in use come from third-party vendors, with the carrier remaining accountable for every regulated decision regardless of origin.

The inventory-first preparation order follows from that accountability. Every artifact an examination requests depends on knowing what AI is actually in use, including the tools embedded in vendor software, and the audit trail design inherits its scope from that inventory. A carrier that documents its governance program before completing its inventory will discover, mid-examination, the systems its records never covered.

The practical synthesis is that the NAIC framework converts the audit trail from an engineering preference into a supervisory expectation: written program, risk-based evaluation, named accountability, and retained documentation, with agentic systems in the highest-impact tier. Carriers should test themselves against the bulletin's evidence question, reconstruct one regulated decision from the retained records, and treat any engineering dependence in that reconstruction as a finding to close before a department raises it.

Frequently asked questions

What does the NAIC framework expect from an insurer's records?

It expects documentation that the governance program operated: a written AIS program, risk-based evaluation of AI systems including vendor-embedded tools, named accountability for outcomes, and records retained for regulatory review. The key distinction is between describing controls and evidencing them, since an examination asks for proof the controls ran. An action-level audit trail across claims, underwriting, and communications is how a carrier produces that proof.

How does state-by-state adoption affect audit trail design?

The adopting state's version governs, and examinations happen where adoption exists, so carriers should build their audit trail to the strictest adopting version they operate under rather than the loosest. More than half of states plus the District of Columbia have adopted the bulletin, and several large states run their own frameworks on the same core expectation. Designing to examination readiness everywhere avoids per-state record inconsistencies.

Do AI agents fall within the bulletin's scope?

Yes, where a state has adopted it: agents acting in claims, underwriting, and customer communications fall within the scope the model describes, and they are usually the highest-impact systems in the carrier's inventory. The risk-based structure expects heightened governance for uses with greater consumer impact, which maps to action-level records and named approvals for consequential agent decisions.

What is the central evidence question in an NAIC-style examination?

Reconstructability: whether the insurer can show one regulated decision, a claim or coverage determination, end to end with approvals and records intact. Carriers should rehearse exactly that reconstruction from retained records alone, because any dependence on engineering effort or missing joins becomes a finding. The preparation order that supports it is inventory first, then program, then evidence.

Sources

  1. content.naic.org
  2. vijil.ai
  3. furtherai.com
  4. behavry.ai
  5. moring.ai
  6. databricks.com
  7. - artificialintelligenceact.eu
  8. - eur-lex.europa.eu