Carrier executives can spend heavily on AI and still misdiagnose why adoption stalls. When a pilot works technically but employees use it inconsistently, hide their use or reject its output, leaders often reach for the same remedies: more training, another mandate or a better tool. Those responses can miss the problem because resistance to AI is not one behavior with one cause.

That matters more in insurance than in many industries. AI can touch underwriting, pricing, claims, fraud detection, customer communication and policy servicing. 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.

A carrier therefore needs more than employee enthusiasm. It needs people to use AI in ways that preserve judgment, surface errors and fit the controls around each workflow. The fastest route to that outcome starts with diagnosing the kind of resistance in front of you.

AI Alarmists Need a Credible Growth Story

Some employees experience AI primarily as a threat. An underwriter may hear “copilot” and translate it into “fewer underwriters.” A claims professional may assume that automation will strip away the judgment and customer contact that give the role meaning. A manager may worry that a successful pilot will eliminate positions on the team.

Telling these employees that AI will “free them for higher-value work” rarely resolves the concern. They have heard similar language during earlier technology rollouts, and they know that efficiency can reduce headcount as easily as it can create capacity. Executives gain more trust by being specific about what growth means.

If the business intends to use AI to increase premium volume without proportional staffing growth, say so. If saved time will be redirected toward complex accounts, broker relationships, catastrophe response, leakage reduction or faster claims resolution, define the work. If some tasks are likely to disappear, acknowledge that and explain how roles, skills and career paths will change.

The practical move is to connect AI adoption to a visible growth commitment. For example, a commercial underwriting team might use AI to assemble submission data and identify missing information while keeping appetite decisions, exceptions and broker negotiations with the underwriter. A claims team might automate document sorting and routine correspondence while reserving coverage disputes, serious injury files and distressed-customer conversations for experienced adjusters. Employees can then see the future role they are being asked to grow into.

Reluctant Adopters Need Safe Visibility

A second group behaves very differently. These employees already see value in AI and may already use it privately, but they keep that use quiet. They may be unsure which tools are approved, what data they can enter, whether managers will judge them for taking shortcuts or whether admitting AI use will make their job look easier than leadership assumed.

This hidden adoption creates two problems: The carrier loses the chance to learn from useful employee experiments, and employees have little incentive to expose mistakes or questionable practices. Shadow AI can spread precisely where governance has the least visibility.

The answer is to create a safe lane for experimentation. Leaders should specify which tools are approved, which information may be entered, where human review is mandatory and how employees can flag a failed experiment without being punished for testing within the rules. Then ask teams to bring forward actual workflows they have already tried.

An underwriter might have found a faster way to summarize inspection reports. A claims examiner might use AI to compare long medical records against a chronology. A compliance analyst might use it to organize regulatory materials before a human review. These are better starting points for adoption than generic tool demonstrations because they reveal where employees already feel enough pain to improvise.

NIST’s Generative AI Profile recommends involving end users and practitioners in prototyping and testing and tracking the outcomes of human-AI configurations over time. That principle fits insurance particularly well. Employees closest to the work often know which exceptions, edge cases and judgment calls a centrally designed pilot will miss.

Pragmatic Resistors Need Verification, Not Persuasion

The third group distrusts AI for practical reasons. They worry about factual errors, biased outputs, weak source data, inconsistent reasoning or a model producing an answer that sounds confident but does not survive scrutiny. In insurance, these concerns deserve respect because a flawed output can affect pricing, coverage, claim handling or regulatory compliance.

Trying to overcome this resistance with enthusiasm is counterproductive. These employees need evidence and a review process they can trust.

Give them explicit verification gates. Define which outputs require source checking, which decisions must remain human, what error thresholds trigger escalation and where the organization records recurring failures. The higher the consumer or financial consequence, the stronger the review standard should become.

For example, an AI-generated summary of a loss run may only require the user to compare key figures against the source document. A recommendation that could influence a declination, pricing exception or disputed claim should face a much higher standard and preserve a documented human decision. The NAIC’s guidance explicitly points insurers toward controls that reflect the nature of the decision, potential consumer harm, human involvement, explainability and reliance on third-party systems.

Pragmatic resistors can become some of the strongest contributors to adoption when leaders use their skepticism well. Ask them to design red-team tests, identify failure modes, define escalation triggers and audit whether review steps catch problems. Their resistance then improves the system instead of sitting outside it.

Middle Managers Turn Policy Into Daily Behavior

Carrier executives can set strategy and approve governance, but employees experience AI through their immediate managers. A written policy may say that experimentation is encouraged, while a manager signals that mistakes will hurt performance reviews. Leadership may announce that human judgment remains essential, while local productivity targets reward speed regardless of review quality.

Middle managers need clear decision rights. They should know which experiments they can approve, which workflows require compliance or legal review, what evidence justifies scaling a pilot, and how to respond when an employee reports a bad AI output. They also need enough practical familiarity with the tools to distinguish reasonable caution from simple avoidance.

This is where many training programs go wrong. Tool instruction without management practice teaches employees how to prompt a system but leaves the social rules unchanged. Carrier leaders should train managers to coach actual workflows, discuss errors without panic and reinforce the review gates that matter most.

Measure Adoption Where Work Changes

License counts and training completion rates tell executives very little about whether AI is improving the business. A better scorecard starts with specific workflows and asks what changed.

For each scaled use case, track a small set of operating measures such as cycle time, error or rework rates, exception frequency, quality review findings, customer outcomes and sustained use. Pair those numbers with evidence about employee behavior. Are people using approved tools rather than private ones? Are they escalating questionable outputs? Are human reviewers catching the errors the control design expected them to catch?

The point is not to maximize AI use. A carrier should want high adoption where the technology improves the workflow and deliberate restraint where the risk exceeds the benefit. The same employee who embraces AI for document summarization may reasonably resist it for a consequential coverage decision.

Executives who recognize these differences can stop treating resistance as a cultural defect. Alarmists need a believable future for their role. Reluctant adopters need permission to bring hidden experiments into the open. Pragmatic resistors need verification standards strong enough to earn trust. Once leaders respond to the reason behind the behavior, AI adoption becomes easier to manage and far easier to govern.