ctuaries have always stood at the intersection of technological innovation, regulatory governance, and legislative oversight. As artificial intelligence transforms core insurance operations, these proficiencies are more crucial than ever. A session at the Casualty Actuarial Society’s recent Ratemaking, Product, and Modeling seminar offered industry perspectives on keeping fairness and governance at the forefront of consumer impacts and company responsibilities regarding AI. The discussion included Jamie Mills, senior actuary at Allstate and session moderator; Will Melofchik, CEO of the National Council of Insurance Legislators (NCOIL); and Jon Godfread, North Dakota Insurance Commissioner.

To level set the discussion, Mills established clear definitions for the various iterations of AI currently impacting the insurance space. While the industry has long utilized data-driven analytics, the rapid emergence of these models requires a shared language to distinguish between their functional capabilities. He identified three AI categories:

• Traditional machine learning: Familiar systems used for modeling and statistical analysis.
• Generative AI: Systems that generate text, summarize documents, and enhance creative work.
• Agentic AI: Systems capable of performing actions, such as interacting with workflows or triggering underwriting steps.

Because legislators often lack a deep insurance background, these categories provide a useful starting point for stakeholders to understand the role of AI in insurance. Bridging this gap is essential to bringing technical innovation to insurers and their customers while ensuring the industry remains committed to fairness, transparency, and accountability.

Human oversight plays a critical role in this process. In one instance within the rental car industry, a series of software glitches led to customers being billed thousands of dollars — a mistake that human intervention in the final review stage could have mitigated.

Melofchik highlighted these concerns among policymakers, noting they are especially focused on material changes or adverse determinations such as policy cancellations, nonrenewals, or significant premium adjustments. He argued that feedback from constituents helps fuel their direction, with headlines about “denial by AI” keeping pressure on legislators to react with new policy. Education on insurance principles like risk-based pricing is critical to helping officials balance insurance challenges against other state priorities such as health care and crime prevention.

While fears surrounding AI’s rapid growth may trigger the impulse to shut down the technology, regulators have also increasingly adopted a view of AI as a powerful tool for an industry that has always relied on sophisticated data analytics. Commissioner Godfread explained how this perspective has translated into actionable regulatory oversight, such as the National Association of Insurance Commissioners (NAIC) Principles on Artificial Intelligence (AI), founded in 2020. These principles prioritize:

• Transparency and explainability: Can a company explain its tools process?
• Safety and integrity: Are company systems secure and are decisions fair?
• Monitoring for bias: Is the company actively checking for unintended bias?

Godfread emphasized that although the tools have evolved, the consumer protection laws foundational to insurance pricing remain unchanged. The ultimate responsibility for a decision lies with the insurance company and its board. He added that the “hardest part” of gaining regulatory approval lies in making complex models understandable. If a model’s output lacks a clear “causation” that makes sense to regulators or the public, it will likely face resistance regardless of its statistical accuracy.

The panelists agreed that continued transparency is needed for building and maintaining “social capital” with both consumers and regulators. Melofchik clarified that most legislators are not seeking access to proprietary code but rather practical transparency, such as informing consumers when they are interacting with an AI chatbot or if AI is driving a nonrenewal decision.

Godfread also noted the importance of transparency within telematics, arguing that while the ability to provide granular risk scores is valuable, the industry must shift the conversation from simple correlation to understandable causation. Similar concerns are growing around aerial imagery and drones, particularly when insurers employ drones or satellite images to non-renew policies due to roof conditions. Legislators are exploring bills that would require insurers to provide these images to consumers and allow a “cure period” (e.g., 60 to 90 days) to resolve the issue before losing coverage to ensure the process remains fair and transparent.

The NAIC’s AI evaluation tool pilot aims to develop mutual transparency between insurers and regulators by standardizing how states review and understand AI usage. Key areas of inquiry include:

• System identification: Categorizing the types of AI systems currently in use across the industry.
• Governance evaluation: Reviewing the oversight mechanisms and structures companies have established.
• Risk management: Understanding how organizations identify and mitigate AI-related risks.

The initiative is in its learning phase, Godfread stressed, as the NAIC actively continues to pursue feedback from insurance professionals on whether the tool is effective without being unnecessarily punitive. Such collaboration will be increasingly vital as fundamental principles of risk, such as risk pooling versus “hyper-personalization,” become more contentious. On this point, Godfread admitted the industry is reaching a point wherein the ability to provide individuals with their exact risk score might conflict with the traditional concept of insurance pools. A solution was noted by Godfread as “TBD,” indicating the issue will require deep intellectual engagement from both regulators and the industry in the coming years.

As states begin to test these AI evaluation tools, Melofchik noted that a primary concern for state-level policymakers is the potential for federal intervention. There is a perceived tension between state legislatures and federal executive orders aimed at creating unified AI standards. Recent legislative activity in states like Utah and Florida has highlighted the delicate balance between state autonomy and the threat of federal preemption. Melofchik explained that many state legislators are wary of federal overreach that might ignore the nuances of the McCarran-Ferguson Act and the historically effective state-based regulatory system. This friction is particularly evident in discussions regarding “human-in-the-loop” mandates.

Commissioner Godfread cautioned that if the insurance industry is lumped into broad, multi-sector federal AI regulations, it could undermine the sophisticated analytics and solvency protections already inherent in the field. The current state-based system already addresses bad actors and technical failures without the need for “one-size-fits-all” federal mandates. For actuaries and executives, this highlights the critical need for active engagement with state legislators to demonstrate that with the existing regulatory structure is capable of evolving alongside AI innovation.

Mills concluded the discussion by addressing concerns AI might replace human professionals, explaining that, as the industry enters an era of complex “black box” models, the need for professional actuarial judgment and a “human touch” becomes more valuable than ever. While repetitive tasks will certainly be automated, the ability to validate a model’s integrity, explain its conclusions, and ensure its ethical application remains a critical human science.

Notably, even “free market” legislators might feel compelled to mandate coverage if the insurance mechanism is perceived as unfair or overly complex, which speaks to an actuary’s role as the critical guardian of model integrity and governance. Ultimately, the ability of actuaries to navigate these issues while maintaining technical accuracy will define the industry’s success in the AI era.