Bản tin mẫu NAIC về AI trong bảo hiểm

The bulletin, formally titled the Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, was adopted by the NAIC in December 2023. The NAIC is an association of state regulators, so its model bulletin has no force until an individual state insurance department issues it, often with light edits. Within about two years, roughly half of US jurisdictions had adopted it or a close version, including Alaska, Connecticut, Illinois, New Hampshire, Pennsylvania and Vermont. Some states took different routes: Colorado has its own statute and regulations on external consumer data and algorithms, and New York's Department of Financial Services issued a separate circular letter in 2024 on AI and external data in underwriting and pricing. The core expectation is an AIS Program: a written program for the responsible use of AI systems, designed to reduce the risk of Adverse Consumer Outcomes, meaning decisions that adversely affect consumers in ways that violate insurance law. The program should cover governance (clear accountability, often up to senior management or the board), risk management and internal controls across the AI lifecycle, data practices, validation and testing, and oversight of third-party models and data. It applies to generative AI as well as predictive models, and it is meant to be proportionate: controls should scale with how much a system affects consumers. A common misconception is that the bulletin bans particular AI uses or sets numerical fairness thresholds. It does neither. It relies on existing statutes such as the Unfair Trade Practices Act and the Unfair Claims Settlement Practices Act, and it describes what documentation regulators may request. Another misconception is that buying a vendor tool shifts responsibility to the vendor. The bulletin is explicit that insurers remain responsible for outcomes from third-party systems they use. For agents, the bulletin is addressed to insurers, but it affects the tools carriers provide to producers and the questions agents should expect about how those tools are used.