Insurers are increasingly using novel data sources and automated systems for risk classification and underwriting. Automation has improved operational efficiencies in the accuracy and speed of underwriting, but it also raises new considerations regarding unfair discrimination. In this paper, we review the current regulatory structures regarding unfair discrimination and suggest they are insufficient to police the myriad new big data sources available. Moreover, artificial intelligence (AI)-enabled systems increase the risk of proxy discrimination, which occurs when a facially neutral factor is utilized as a proxy for a prohibited characteristic. Furthermore, AI-enabled systems raise unique ethical implications, particularly regarding accountability among AI actors. Many insurers rely on unregulated third-party algorithm developers; therefore, they do not own and may not have access to the logic embedded in the system. 

To address these issues, we propose a framework that consists of three parts: 1) the establishment of national standards to serve as guardrails for the acceptable design and behavior of AI-enabled systems; 2) a certification system that attests that an AI-enabled system was developed in accordance with those standards; and 3) periodic audits of the systems’ output to ensure it operated consistent with those standards. The framework rests on the existing state-based regulatory infrastructure and envisions a self-regulatory organization that can work with the NAIC to develop standards and oversee certification and audit processes. Regulatory enforcement remains with the states.