Why Insurance AI Rollouts Stall — and Why the NAIC Is Now Watching

Why Carrier AI Pilots Work on Paper but Stall in Practice

Insurance carriers are pouring money into artificial intelligence, and in many cases the technology works. The problem is what happens next: employees use the tools inconsistently, quietly work around them, or reject the output altogether. When that happens, executives tend to reach for the same three remedies — more training, another mandate, or a better tool. According to the source article, those responses often miss the real issue, because resistance to AI is not a single behavior with a single cause.

That distinction carries more weight in insurance than in many other industries. AI can touch underwriting, pricing, claims, fraud detection, customer communication and policy servicing — functions where decisions directly affect consumers and are already heavily regulated. A model that performs well in a pilot can still create compliance exposure if staff use it unevenly or override it without documentation.

The regulatory backdrop is tightening. The NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers makes clear that consumer-impacting decisions supported by AI remain subject to existing insurance laws and regulatory standards, and it expects insurers to maintain governance and risk controls around those systems. In 2026, the NAIC is also piloting an AI Systems Evaluation Tool with 12 states to help regulators examine insurer governance, risk mitigation, data inputs and potentially high-risk AI models.

The practical implication is that adoption failure is no longer just an internal efficiency problem. If employees use AI inconsistently, the gap between what a carrier's governance framework claims and what actually happens on the desk becomes a regulatory exposure — one that a 12-state examination pilot is explicitly designed to surface.

What the NAIC Pilot Means for Carrier AI Governance

Why 'More Training' Is the Wrong Diagnosis

The source article's central argument is that resistance to AI is not one behavior. An employee who hides their use of a tool is behaving differently from one who rejects its output outright, and differently again from one who uses it inconsistently. Each pattern points to a different root cause — fear of being replaced or penalized, distrust of the model's accuracy, or simple friction in the workflow. A single remedy applied across all three will fail at least two of them. For carriers, the cost of misdiagnosis is not just wasted training budget; it is a rollout that never reaches the consistency regulators will expect.

Why Insurance Is a Harder Case Than Most Industries

In most sectors, a stalled AI rollout is an efficiency problem. In insurance, it is also a compliance problem. AI touches underwriting, pricing, claims, fraud detection, customer communication and policy servicing — decisions that affect consumers and are governed by existing insurance law. The NAIC Model Bulletin explicitly states that AI-supported consumer-impacting decisions remain subject to those laws and standards, and expects insurers to maintain governance and risk controls around the systems. That means inconsistent employee use is not merely an adoption metric; it is a gap between documented governance and actual practice.

What the 2026 NAIC Pilot Changes

The NAIC's AI Systems Evaluation Tool pilot with 12 states is designed to help regulators examine insurer governance, risk mitigation, data inputs and potentially high-risk AI models. For context, this fits a broader pattern of US financial regulators moving from principle-based guidance toward examination-based oversight of AI. The practical consequence for carriers is that the questions regulators ask will increasingly be about what happens on the desk, not just what is written in the policy. A carrier that cannot demonstrate consistent, documented use of an AI system may struggle to show that its governance framework is functioning as intended.

Where This Leaves Carrier Executives

The carriers best positioned for the 2026 pilot are those that treat adoption diagnostics as part of their AI governance, not separate from it. That means distinguishing between the three resistance types, measuring them separately, and connecting the findings to the risk controls the NAIC expects. Carriers that continue to respond to every adoption problem with more training or another mandate risk arriving at the examination with a governance framework that looks complete on paper but does not match how employees actually use the tools.

How Carrier Executives Should Respond Before 2026

Adoption failure in insurance AI is a governance risk, not just a training gap — and the NAIC's 2026 pilot will test whether carriers can prove it.

For carrier executives, the source article's core insight is that resistance to AI has at least three distinct forms — inconsistent use, hidden use and outright rejection of output — and each requires a different response. Applying one remedy across all three wastes budget and leaves the underlying governance gap unaddressed.

• Diagnose before you train. Separate the three resistance types in your own rollout data. An employee hiding their use of an AI tool is signaling something different from one who rejects its output, and the fix is not the same.
• Map AI use to the NAIC Model Bulletin's scope. Identify which AI systems touch consumer-impacting decisions in underwriting, pricing, claims, fraud detection, customer communication and policy servicing — those are the systems the Model Bulletin covers and the ones the 2026 pilot is designed to examine.
• Close the gap between documented governance and desk-level practice. If your governance framework assumes consistent AI use but employees use tools unevenly, that gap is the exposure. Document actual usage patterns, not intended ones.
• Prepare for the 12-state pilot's focus areas. The NAIC's AI Systems Evaluation Tool will examine governance, risk mitigation, data inputs and potentially high-risk models. Carriers should be able to demonstrate controls in each of those areas before examination requests arrive.

Metrics worth watching: the NAIC's rollout schedule and scope for the 12-state AI Systems Evaluation Tool in 2026, and any state-level examination activity referencing the Model Bulletin.