Book a demo
Topics · COST-BUDGET

Insurance Agent Cost NAIC Bulletin | Moring

The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted December 4, 2023, expects a written AIS Program with governance, risk management, and internal controls proportionate to consumer risk. Agent cost observability supplies part of that program: attribution, runtime caps, and retained budget events as dated evidence for market conduct examination.

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, adopted December 4, 2023, expects a written AIS Program with governance, risk management, and internal controls proportionate to consumer risk. Agent cost observability supplies part of that program: attribution, runtime caps, and retained budget events as dated evidence for market conduct examination.

How it works

The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers is the anchor document for insurance AI governance in the United States. Adopted December 4, 2023, it does not create a new statute; it tells insurers how regulators intend to use existing unfair trade practices statutes and market conduct examination authority to review how AI systems are developed, acquired, and used where they make or support consumer-impacting decisions. Most states have since adopted it as guidance.

The bulletin's central expectation is a written AI System Program, an AIS Program, covering governance, risk management, and internal controls. Its controls are expected to be proportionate to the risk an AI system poses to consumers, and its scope reaches underwriting, rating, claims handling, marketing, and fraud detection, which covers nearly every production use case a carrier operates. An agent estate therefore lives inside the bulletin's perimeter by default.

Cost observability enters the program as an internal control with an evidentiary product. The attribution record, stamping every model call and tool invocation with the agent identity, workflow, and owning team, demonstrates which AI systems were operating, at what scale, and under whose ownership. The runtime budget, warning at soft-cap thresholds and refusing at hard limits, demonstrates a control that operates continuously rather than a policy that exists on paper.

The bulletin's examination logic is what makes the retained record decisive. Compliance analysis of the bulletin emphasizes that insurers must be able to produce dated evidence for specific systems, and research on agentic AI in insurance adds the operational constraint: traces, guardrail decisions, and override logs must be captured as the agent acts, because they cannot be rebuilt when the exam letter arrives. Budget events, refusal records, and human fallback decisions belong in that same captured-as-acting category.

Proportionality guides the configuration. A pricing or underwriting model that directly affects policy terms warrants different scrutiny than an internal fraud-detection tool, and the same scaling applies to cost controls: tighter caps, deeper attribution, and more frequent review for agents whose actions reach consumers. Configuring cost observability per risk tier keeps the AIS Program coherent rather than a pile of undifferentiated controls.

Vendor accountability is an explicit bulletin theme. Insurers remain accountable for the compliance of AI systems and predictive models obtained from vendors, including monitoring for unfair discrimination, which means the cost and usage of vendor-supplied agents should be observable on the same terms as internal systems: identity-attributed, budget-capped, and retained. A governance gap in a vendor's platform becomes the carrier's exposure, not the vendor's.

The bulletin also reaches decisions or actions impacting consumers made or supported by advanced analytical technologies, which includes agentic systems, and multi-state adoption continues to broaden its practical reach. For a carrier, the cost observability program should be documented inside the AIS Program: the attribution scheme, the cap design, the escalation owners, and the retention policy, each mapped to the internal-control expectation it satisfies.

Frequently asked questions

What does the NAIC Model Bulletin actually require of insurers?

It expects each insurer to maintain a written AI System Program covering governance, risk management, and internal controls, proportionate to the risk each system poses to consumers, and it reaches underwriting, rating, claims handling, marketing, and fraud detection. It does not create new law; it signals how existing unfair trade practices statutes and market conduct examination authority will be applied. Adopted December 4, 2023, it has since been taken up as guidance by a majority of states.

How does cost observability fit inside an AIS Program?

As an internal control with an evidentiary product for the insurer: attribution stamps every model call and tool invocation with the agent identity, workflow, and owning team, and runtime budgets warn at soft-cap thresholds and refuse at hard limits. The retained record of warnings, refusals, and human fallback decisions demonstrates the control operating continuously, which supports the bulletin's dated-evidence expectations during market conduct examination.

Why must cost evidence be captured as the agent acts?

Because traces, guardrail decisions, override logs, and budget events cannot be rebuilt when the exam letter arrives; retrofitting them from invoices and memory is neither complete nor credible. Append-only storage with per-agent identity attribution seals the record as it accumulates, so the insurer's examination answers come from retained evidence of what each system was allowed to do and what it did.

How does the bulletin treat vendor-supplied agents from a cost standpoint?

Insurers remain accountable for the compliance of AI systems obtained from vendors, so vendor-supplied agents should be observable on the same terms as internal ones: identity-attributed, budget-capped, and retained. A governance gap in a vendor's platform becomes the carrier's regulatory exposure, which makes the ability to show what vendor agents cost and how their spend was monitored part of the AIS Program rather than an optional extra.

Sources

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