How the NAIC AI Bulletin Applies to Workers' Compensation Claims AI

What the Bulletin expects of an insurer

The NAIC adopted the Model Bulletin on December 4, 2023. A model has no direct legal effect until a state issues its own bulletin or equivalent, so the obligations an insurer faces come from the bulletin each state has issued. The model nonetheless sets the shared vocabulary that examiners use.

The Bulletin reminds insurers holding certificates of authority that decisions or actions impacting consumers, when made or supported by advanced analytical and computational technologies including AI systems, must comply with all applicable insurance laws and regulations. It describes how insurers should govern the development, acquisition, and use of certain AI technologies, and it tells insurers what information and documentation a department may request during an investigation or examination.

Four expectations matter most for claims. First, a written AI system (AIS) program designed to mitigate the risk of adverse consumer outcomes. Second, risk management and internal controls around that program. Third, oversight of AI that a third party develops or operates. Fourth, the ability to produce documentation to a regulator on request. NAIC commentary on the Bulletin describes coverage across the insurance life cycle, including case management, claim administration and payment, and fraud detection. Claims is squarely in view.

The phrase "made or supported by" is the practical scope question. It suggests that AI informing a human adjuster's decision can fall within the Bulletin, not only fully automated outcomes. Compliance teams should read each state's adopted language on this point rather than assume the model text carries over unchanged.

Mapping workers' comp claims AI use cases to the Bulletin

The NAIC AI Systems Evaluation Tool lists claims examples that include claim assignment, triage and fast-tracking, individual and bulk reserving including loss estimation, imaging and video analysis, fraud detection, litigation, estimation of closure rates, salvage and subrogation, and examination or report ordering. Several of these overlap directly with common workers' compensation uses. Others are not named, so scope depends on how the output is used in a claims decision.

For each use case, the practical question is the same: does the output influence a decision or action affecting a claimant, employer, or provider? Where it does, plan for documentation of purpose, data inputs, testing, and the point at which a person reviews the output.

• Triage and assignment: Named in the evaluation tool. Likely in scope because routing and fast-tracking shape how a claim is handled. Document the criteria and where adjusters can override.
• Severity scoring: Overlaps with the tool's loss estimation and closure-rate examples. Document model purpose, testing, thresholds, and how adjusters use the score.
• Reserving support: Named in the tool as individual and bulk claim reserving. Record who approves reserve changes and whether the AI output was a factor.
• Fraud indicators: Fraud detection is named in the Bulletin context and the tool. Define how an indicator is used and what human review precedes any adverse action.
• Medical bill review: Not named in the tool excerpts. Assess against the made-or-supported-by language, especially if outputs drive payment or reduction decisions.
• Claims-note summarization: Not named. Scope turns on whether summaries inform decisions. Even if lower risk, data handling and accuracy controls still apply.

What the Bulletin requires versus what states and claims law add

Keeping these layers separate prevents both over-claiming and gaps. The Bulletin sets expectations for the program, controls, third-party oversight, and documentation. It does not replace the statutory and claims-handling obligations that workers' compensation already carries, which differ by state and apply whether or not AI is involved.

Adoption of the Bulletin is also state by state. The NAIC implementation map lists state bulletins from Alaska (February 1, 2024) through Wisconsin (March 18, 2025), including Iowa, Massachusetts, North Carolina, and Oklahoma in late 2024 and Delaware and New Jersey in February 2025. As of October 2026, adoption remains a matter of each state's own bulletin, and states may vary in wording.

For a carrier or TPA operating in several states, this argues for a simple record: for each claims AI use case, which state bulletin governs, which workers' compensation claims-handling duties apply, and which control or document addresses each. That record turns a general program into something an examiner can follow.

Third-party AI deserves its own line in that record. Many claims AI capabilities sit inside vendor or TPA platforms. The insurer remains responsible for oversight, so the question is what documentation and per-claim records each vendor can actually supply when a regulator asks.

Where Trussed fits

Trussed is control, provenance, and evidence infrastructure for consequential AI. It enforces policy in the execution path across models, agents, tools, and MCP servers on individual interactions and tool calls: allow, redact, block, rate limit, or escalate for human approval. Policy is enforced at runtime, not reviewed after the fact.

That enforcement generates continuous audit-ready evidence mapped to NAIC, among other frameworks, which gives a claims organization a per-interaction record of what its AI was permitted to do and what happened. Deployment is self-managed in the customer cloud, with configurable retention and data residency. Trussed runs in production for customers in insurance and other regulated verticals.

It does not replace model validation, bias testing, vendor due diligence, or the written AIS program. Those remain governance work, and the runtime evidence is meant to sit alongside them.