Book a demo
Topics · GOVERNANCE-AGENTS

Insurance Audit Trail NAIC Bulletin | Moring

The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers defines what an insurer must document: a written AIS program owned by senior management accountable to the board, risk-based evaluation across owned and vendor systems, and records retained for investigation or examination. Its Section 4 enumerates what a department may request, which makes the audit trail the compliance artifact.

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

Short answer

The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers defines what an insurer must document: a written AIS program owned by senior management accountable to the board, risk-based evaluation across owned and vendor systems, and records retained for investigation or examination. Its Section 4 enumerates what a department may request, which makes the audit trail the compliance artifact.

How it works

The NAIC adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers on 4 December 2023, and it remains the reference text for how state regulators expect carriers to govern AI, including agentic systems. The bulletin directs insurers to develop, implement, and maintain a written program for the responsible use of AI systems, with responsibility sitting with senior management accountable to the board or an appropriate committee. That accountability placement moves AI governance into the reporting structure that already carries corporate governance annual disclosure obligations, because the bulletin is explicit that those disclosure requirements reach any part of the governance framework addressing AI-supported decisions that affect consumers.

Section 4 is the operational heart for audit trail purposes: it lists the type of information and documentation a department may request during an investigation or examination of an insurer's AI use. Governance documentation, evaluation records, testing methodology, and vendor oversight artifacts are all in scope, and the practical reading is that anything the program describes should be producible as evidence. A policy statement describing governance intentions does not satisfy the requirement; a documented program with audit trails does.

The bulletin's structure is risk-based. AI uses with greater potential to adversely affect consumers are expected to receive heightened governance controls, while lower-impact uses may warrant lighter oversight. For carriers, that tiering translates directly into record depth: agent actions touching claim outcomes, coverage determinations, or pricing need complete action-level records with named human accountability, while lower-risk automation needs demonstrable but lighter evidence. The tiering decision itself should be documented, since it is the first thing that determines what the trail must contain.

Scope covers both predictive models and generative AI systems used across underwriting, rating, claims handling, marketing, and fraud detection, and it applies to third-party and vendor-embedded systems, with the carrier remaining accountable. Adoption surveys show why vendor scope matters: a large share of models in insurer use come from outside vendors, yet regulated decisions remain the carrier's responsibility, so vendor agent behavior needs records the insurer can produce, not records the vendor keeps.

Enforcement posture is currently exam-readiness rather than imminent penalty. No formal enforcement action under an adopted bulletin had been publicly reported as of early 2026, but the examination infrastructure is active: the AI Systems Evaluation Tool, renamed the AI Risk Evaluation Supplement, began piloting with twelve states in March 2026. The pressure is therefore on documentation quality and reconstructability, which is precisely what an audit trail provides or fails to provide.

Carriers aligning with the bulletin should treat the audit trail as the implementation of its documentation expectation: one structured record design across claims, underwriting, and rating, bound to the case, protected from tampering, retained per the insurer's schedule, and capable of answering the Section 4 request list as a query rather than a project.

Frequently asked questions

What does Section 4 of the NAIC bulletin mean for audit trails?

Section 4 lists the information and documentation a department may request during an investigation or examination of an insurer's AI use, which effectively defines the audit trail's deliverables: governance documentation, evaluation and testing records, and vendor oversight artifacts. Anything the AIS program describes should be producible as evidence on request. Carriers should map their record design against that list and close gaps before an inquiry arrives.

Who is accountable for the AIS program under the bulletin?

Senior management is responsible for developing, implementing, monitoring, and overseeing the program, accountable to the board or an appropriate committee, which places AI governance inside the structure already carrying corporate governance annual disclosure obligations. For the audit trail, that means evidence of operation must reach board-level reporting, not remain an engineering artifact. Named accountability extends to individual AI systems and their outcomes.

Does the bulletin cover vendor and generative AI systems?

Yes. It applies across predictive models and generative AI used in underwriting, rating, claims handling, marketing, and fraud detection, including third-party and vendor-embedded systems, with the carrier remaining accountable for regulated decisions regardless of origin. Given the large share of insurer models sourced from vendors, records of vendor agent behavior must be producible by the carrier, which shapes contract and logging requirements.

What is the current enforcement posture around the bulletin?

No formal enforcement action under an adopted bulletin had been publicly reported as of early 2026, but examination infrastructure is active, including the piloted AI Systems Evaluation Tool, renamed the AI Risk Evaluation Supplement, with twelve participating states. The practical pressure is exam readiness: documentation quality and the ability to reconstruct regulated insurance decisions. Carriers should treat this window as the time to build records rather than respond to findings.

Sources

  1. content.naic.org
  2. vijil.ai
  3. trussed.ai
  4. behavry.ai
  5. optro.ai
  6. - artificialintelligenceact.eu
  7. - eur-lex.europa.eu