NIST AI RMF and the NAIC AI Model Bulletin: What Alignment Covers and What Evidence Remains

What the bulletin permits and what it expects

The model bulletin sets out the Department's expectations for how insurers govern the development, acquisition and use of AI Systems. It states that an AIS Program may adopt, incorporate or rely upon, in whole or in part, a framework from an official third-party standard organization, and it names the NIST AI Risk Management Framework, Version 1.0, as an example.

Permission to rely on a framework is not a finding that the framework satisfies the bulletin. The bulletin's underlying premise is that decisions or actions affecting consumers, made or supported by AI Systems, must comply with all applicable insurance laws and regulations. The program is meant to mitigate the risk of Adverse Consumer Outcomes and to cover the insurance life cycle, including product development and design, marketing, underwriting, rating and pricing, case management, claim administration and payment, and fraud detection.

That scope matters for evidence. Every AI system touching those areas, whether built in-house or acquired from a vendor, is a system for which the insurer needs to be able to show governance and control.

Crosswalk: RMF functions and AIS Program expectations

The crosswalk below is conceptual. It shows where each RMF function supports an AIS Program element and what the framework alone leaves undone. Confirm specific provisions against the bulletin your state adopted.

• Govern: Supports: AIS Program governance, accountability and documented legal and regulatory requirements. Remaining: a written program tied to the bulletin, plus records showing the governance was applied to named systems.
• Map: Supports: understanding each system's context, purpose and impact. Remaining: an inventory covering every system in the life-cycle areas, including third-party systems, linked to decision types.
• Measure: Supports: assessing and tracking risk. Remaining: retained results that show how risk was evaluated for consequential decisions, and ongoing records rather than point-in-time reviews.
• Manage: Supports: prioritizing and responding to risk. Remaining: internal controls that act on the system and records of vendor oversight, showing the response occurred and not only that it was planned.

Policy documentation versus runtime operating evidence

GOVERN 1.1 states that legal and regulatory requirements involving AI are understood, managed and documented. That is a policy-level outcome. A well-maintained set of AIS Program documents can satisfy it and still say nothing about what a particular model did on a particular claim or underwriting decision last month.

Examiners working from a written program tend to test whether the program is operating. The difference is practical. A policy says personal data should not be sent to an external model. An operating record shows that a given request was evaluated against that rule, what the outcome was, and whether a person approved an exception.

Third-party AI sharpens the gap. Contract terms and vendor diligence are documents. Evidence of oversight is a record of how the vendor's system behaved inside your workflows and which controls applied to it. If a control exists only as a procedure that people are expected to follow, the insurer can show intent but not necessarily operation.

The bulletin also contemplates regulator information requests. Plan for them by treating retained, system-specific records as part of the AIS Program rather than something assembled after a request arrives.

State variation and parallel regimes

The model bulletin is adopted state by state, and each jurisdiction's own bulletin governs. As of October 2026, the NAIC adoption map, last updated in August 2026, lists 21 jurisdictions: Alaska, Arkansas, Connecticut, Delaware, the District of Columbia, Illinois, Iowa, Kentucky, Maryland, Massachusetts, Michigan, Nevada, New Hampshire, New Jersey, North Carolina, Oklahoma, Rhode Island, Vermont, Virginia, West Virginia and Wisconsin. Listed adoption dates run from February 2024 to March 2025.

A single RMF-based program therefore needs a state check. For each state where you write or broker business, compare its bulletin with the model text for differences, and confirm status directly for states not on the list.

Other regimes can apply alongside the bulletin. NYDFS Circular Letter 7 is relevant to carriers in New York. Colorado SB 24-205 addresses algorithmic discrimination in consequential decisions made by high-risk AI systems, and it requires developers to disclose known or reasonably foreseeable discrimination risks to the attorney general and to known deployers or other developers within 90 days of discovery or of a credible deployer report. An RMF-based AIS Program should not be assumed to cover duties such as these without a separate review.

Where runtime enforcement fits

Closing the evidence gap requires controls that act where the AI acts. Trussed enforces policy in the execution path across models, agents, tools, MCP servers, AI sites and coding assistants. Each interaction or tool call can be allowed, redacted, blocked, rate limited or escalated for human approval. Policy is enforced at runtime, not reviewed after the fact.

That enforcement generates audit-ready evidence continuously, mapped to NAIC, NYDFS and NIST requirements. The evidence records what the AI was allowed to do and what it did, which is the material a documented AIS Program otherwise lacks. Trussed can be self-managed in the customer's cloud, with configurable retention and data residency.

It operates in production today, with 18+ months of continuous production in a regulated enterprise, 99.95% uptime on live production traffic and 100+ models under one policy plane. Customers span four regulated verticals, including insurance, and Trussed holds NIST AI RMF alignment among its certifications. Enforcement scope is individual interactions and tool calls. It supplements the written program, the inventory and the risk assessment work, and does not replace them.