Adoption of NAIC Model Bulletin

Twenty states and districts have adopted the NAIC Model Bulletin on the use of AI by insurers, in part or in whole, as of: December 11, 2024.

In order of adoption:

• Alaska, February 1, 2024, Bulletin B 24-01: No notable changes

• New Hampshire, February 20, 2024, Docket #INS 24-011-AB:

  • Strongly encourages development of verification and testing methods to identify errors and limit unfair discrimination (as opposed to simply encourages)

  • Updates “bias analysis and minimization” in reference to data management practices to “unfair bias analysis and minimization”

• Nevada, February 23, 2024, Bulletin 24-001: No notable changes

• Connecticut, February 26, 2024, Bulletin MC-25: 

  • Insurers must complete annual Artificial Intelligence Certification that attests to compliance with the Bulletin

• Vermont, March 12, 2024, Insurance Bulletin 229: No notable changes

• Illinois, March 13, 2024, Company Bulletin 2024-08: No notable changes

• Rhode Island, March 15, 2024, Insurance Bulletin 2024-03: No notable changes

• Pennsylvania, April 6, 2024, Notice 2024-04: No notable changes

• Kentucky, April 16, 2024, Bulletin 2024-02: 

  • Clarifies that the guidelines in the bulletin are non-binding and softens the overall tone, using phrases like “the Department recommends,” “the Department advises,” or “controls would ideally address” 

• Maryland, April 22, 2024, Bulletin 24-11:

  • Extends scope from insurers to include nonprofit health service plans, HMOs, and dental plan organizations

• Washington, April 22, 2024, Technical Assistance Advisory 2024-02: No notable changes

• District of Columbia, May 21, 2024, Bulletin 24-IB-002-05/21: No notable changes

• Nebraska, June 11, 2024, IGD -- H1: No notable changes

• Virginia, July 22, 2024, Administrative Letter 2024-01:

  • Strongly encourages development of verification and testing methods to identify errors and limit unfair discrimination (as opposed to simply encourages)

  • Updates goal of program from “mitigate the risk” of adverse consumer outcomes to “eliminate the risk”

  • Strengthens much of the language, such as AI governance programs “should address” the various items listed as opposed to “consider addressing”

• Arkansas, July 31, 2024, Bulletin 13-2024: No notable changes

• Michigan, August 7, 2024, Bulletin 2024-20-INS: No notable changes

• West Virginia, August 9, 2024, Insurance Bulletin 24-06: Notable changes

  • Softens language from what insurers “should” do in their AI governance programs to what “OIC … recommends” the insurers “would ideally” do

  • Largely removes sections on legislative authority as well as regulatory oversight and examination considerations, while noting that an investigation or market conduct action may include questions about use of AI and related governance

• Iowa, November 7, 2024, Bulletin 24-04: Notable changes

  • First state to define “bias,” which the Division encourages insurers to identify and manage: a distortion or error in statistical analysis that produces inaccurate results. An AI System may be said to be biased when it exhibits systematically inaccurate behavior

  • Also defines “outcomes testing” in relation to unfair discrimination: validating that a model or AI system is working as intended and does not produce unfairly discriminatory outcomes 

  • Notes that the Division anticipates providing additional guidance regarding the governance of third-party AI Systems

• Oklahoma, November 14, 2024, Bulletin No. 2024-11: No notable changes 

• Massachusetts, December 9, 2024, Bulletin 2024-10: No notable changes 

In addition, Colorado and New York have adopted their own frameworks for regulating AI use by insurers.