Artificial Intelligence in Insurance Claims: What Policyholders Need to Know

The insurance industry has embraced artificial intelligence at every stage of the claims process. From the moment a claim is reported to the final settlement offer, algorithms now influence — and in some cases entirely determine — how much a policyholder receives. Insurers describe this transformation as faster, more consistent, and more efficient. What they rarely discuss is whether it is more accurate or more fair.

Policyholders need to understand how these systems work, where they fail, and what rights exist to challenge AI-driven decisions. The speed of AI adoption has far outpaced the regulatory framework governing its use, and the consequences fall disproportionately on the people filing claims.

How Insurers Are Using AI in Claims Handling

Artificial intelligence is not a single technology. In insurance claims, it encompasses a range of tools deployed at different stages of the process:

Automated Triage and Claims Routing

When a claim is first reported, AI systems now classify it by severity, complexity, and expected cost. Simple claims — a broken window, a minor kitchen fire, a small water leak — may be routed to fast-track processing with minimal human involvement. More complex claims are categorized and assigned to adjusters, but the AI’s initial classification can set the tone for the entire handling process. A claim flagged as “low severity” by the algorithm may receive less attention, fewer resources, and lower initial reserves than it deserves.

Photo-Based Damage Estimation

Several major insurers now use computer vision systems that analyze photographs of damage to generate repair estimates. A policyholder submits photos of a damaged roof, and within minutes the AI produces a scope of work and dollar figure. These systems are trained on historical claims data — millions of images paired with the amounts insurers previously paid. The problem is that the training data reflects what insurers chose to pay, not what repairs actually cost. If the historical data systematically undervalues claims, the AI learns to undervalue them as well.

Fraud Scoring

AI-driven fraud detection models assign risk scores to incoming claims based on dozens of variables: the timing of the loss, the policyholder’s claim history, the amount claimed relative to the policy limits, geographic patterns, and more. A high fraud score can trigger a Special Investigations Unit (SIU) referral, which dramatically slows the claims process and subjects the policyholder to intensive scrutiny. The criteria for these scores are proprietary and opaque. A policyholder flagged as “high risk” may never know why, and may face significant delays as a result.

Settlement Recommendations

Perhaps the most consequential application of AI in claims is the generation of settlement recommendations. These systems analyze the claim data, comparable claims, policy terms, and sometimes even the policyholder’s likelihood of hiring a lawyer or filing a complaint, to produce a recommended payout amount. The adjuster assigned to the claim may have limited authority to deviate from the algorithm’s recommendation. In effect, the AI sets the initial offer, and the human adjuster becomes the messenger.

The Speed Promise — and Its Cost

Insurers frequently tout the speed gains of AI-driven claims handling. Claims that once required ten days of adjuster investigation, contractor coordination, and desk review can now close in as little as 36 hours. For straightforward claims with clear coverage and accurate damage assessment, this speed is genuinely beneficial. No policyholder wants to wait weeks for a simple repair payment.

The danger is that speed and accuracy are not the same thing. A claim closed in 36 hours is not necessarily a claim that was handled correctly. If the AI underestimated the damage, misclassified the cause of loss, or applied an exclusion that does not actually apply, the fast closure simply means the policyholder was underpaid faster. And the speed itself creates pressure to accept: when an insurer presents a settlement offer within days, policyholders may feel compelled to take it before fully understanding the extent of their loss.

The Accuracy Problem: AI Trained on Insurer Data

Every machine learning system is only as good as the data it was trained on. Insurance AI systems are trained on historical claims data — data generated by the insurers themselves. This creates a fundamental problem: if insurers have historically undervalued claims, the AI will learn to undervalue them. If desk reviewers have routinely reduced field adjuster estimates, the AI will learn that reduced amounts are “correct.” If certain types of damage have been systematically excluded or minimized, the AI will replicate those patterns.

This is not a theoretical concern. The insurance industry has a well-documented history of claims suppression practices. Programs that instruct adjusters to reduce estimates, deny certain claim types at higher rates, or apply aggressive interpretations of policy exclusions have been exposed repeatedly through litigation and regulatory action. When AI systems are trained on the output of those practices, they do not correct the bias — they encode it.

The result is a system that appears objective and data-driven but systematically produces outcomes favorable to the insurer. A policyholder dealing with an adjuster who says “the system calculated your payout” is facing an opponent that has dressed up historical underpayment as algorithmic precision.

The Regulatory Landscape: NAIC and State Action

The National Association of Insurance Commissioners (NAIC) has recognized the risks of AI in insurance. In late 2023, the NAIC adopted a Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, which established a governance framework for AI use across the industry. The bulletin requires insurers to implement AI governance programs, conduct risk assessments, and ensure that AI-driven decisions comply with existing unfair trade practice laws.

By late 2025, more than 24 states had adopted or were in the process of adopting the NAIC Model Bulletin or substantially similar guidance. This represents a significant shift in regulatory attention, though the Model Bulletin is principles-based rather than prescriptive, meaning enforcement varies significantly by state.

The core requirements of the NAIC framework include:

• AI governance programs. Insurers must establish internal programs to oversee the development, deployment, and monitoring of AI systems used in underwriting, rating, and claims.
• Risk assessment. Before deploying AI in claims handling, insurers must assess the risk that the system could produce unfairly discriminatory outcomes or violate existing consumer protection laws.
• Compliance with existing law. The bulletin makes clear that existing unfair claims settlement practices statutes apply fully to AI-driven decisions. An insurer cannot use an algorithm to do what it would be prohibited from doing manually.
• Human oversight. The framework emphasizes that AI should augment, not replace, human decision-making in consequential claims determinations.

California SB-1120: A Model for AI Accountability

California has been at the forefront of AI regulation in insurance. SB-1120, signed into law as part of the state’s broader consumer protection framework, establishes that artificial intelligence cannot be the sole decision-maker in certain insurance contexts. The law requires meaningful human review of AI-generated decisions that adversely affect policyholders, including claim denials, significant coverage reductions, and policy non-renewals.

This is a critical protection. It means that when an insurer denies a claim or offers a significantly reduced settlement, a human being must review the AI’s recommendation before the decision is communicated to the policyholder. The human reviewer must have sufficient expertise and authority to override the algorithm’s output. A rubber-stamp review that simply approves whatever the AI recommends does not satisfy the requirement.

Court Rulings: Discovery Into AI Use

Courts are increasingly willing to allow discovery into how insurers use AI to make claims decisions. In multiple jurisdictions, judges have permitted policyholders and their attorneys to obtain information about the algorithms, models, and scoring systems used to evaluate and deny claims. This is significant because insurers have historically treated their claims technology as proprietary trade secrets, shielded from outside scrutiny.

The legal reasoning is straightforward: when an insurer relies on an AI system to deny or reduce a claim, the policyholder has a right to understand how that decision was made. In bad faith litigation, the reasonableness of the insurer’s conduct is the central question. If the insurer delegated that conduct to an algorithm, the algorithm’s design, training data, and outputs become directly relevant to whether the insurer acted reasonably.

Discovery into AI use has revealed troubling patterns in several cases: models that were never validated against actual repair costs, systems that flagged claims from certain geographic areas at higher rates, and algorithms that recommended denials based on the policyholder’s estimated likelihood of pursuing litigation rather than the merits of the claim itself.

Policyholder Rights When AI Is Involved

The growing use of AI in claims handling does not diminish policyholder rights. Every protection that applies to human-made claims decisions applies equally to AI-driven ones. In addition, the emerging regulatory framework creates new rights specific to algorithmic decision-making:

The Right to Know If AI Was Used

Policyholders should ask — in writing — whether any AI, algorithmic, or automated decision-making system was used in evaluating their claim. This includes automated damage estimation tools, fraud scoring models, and settlement recommendation engines. In states that have adopted the NAIC framework or similar regulations, insurers are expected to be transparent about the use of AI in claims decisions. Even where disclosure is not yet explicitly required, asking the question creates a record that can be valuable in any subsequent dispute.

The Right to Challenge AI-Driven Denials

An insurer cannot deny a claim simply because an algorithm recommended denial. The insurer must still have a reasonable basis for the denial under the policy terms, supported by an adequate investigation. If the denial is based on an AI assessment that the policyholder believes is inaccurate — for example, a photo-based damage estimate that missed significant damage, or a fraud score triggered by innocent circumstances — the policyholder has every right to challenge it.

The challenge should be specific and documented. If the AI underestimated the damage, provide independent contractor estimates, photographs, and expert assessments that demonstrate the actual scope of loss. If the AI misclassified the cause of loss, present evidence of the actual cause. Treat an AI-generated denial the same way as any other denial: demand a written explanation, gather supporting evidence, and escalate if the insurer refuses to reconsider.

The Right to Human Review

In California and an increasing number of other jurisdictions, policyholders have the right to request that a qualified human adjuster review any AI-generated claims decision. This is not a request for a second opinion from the same algorithm — it is a request for a human being with subject matter expertise to independently evaluate the claim. If the human reviewer simply defers to the AI output without conducting an independent analysis, the review is inadequate and may support a bad faith claim.

What to Do If You Suspect an AI-Driven Denial

If a claim has been denied, underpaid, or unreasonably delayed and the policyholder suspects AI was involved, the following steps can help protect policyholder rights:

• Request written confirmation of AI use. Ask the insurer, in writing, whether any automated or algorithmic tools were used in evaluating the claim, estimating damages, scoring for fraud, or generating the settlement offer.
• Request a detailed written explanation of the denial. The insurer must explain the specific reasons for the denial or reduced payment. If the explanation references software-generated estimates or system-calculated amounts, note this in any response.
• Obtain independent estimates.Do not rely solely on the insurer’s damage assessment. Obtain independent contractor estimates or, for larger claims, retain a licensed public adjuster to prepare an independent scope and estimate.
• Document the discrepancy.If the insurer’s AI-generated estimate is significantly lower than independent estimates, document the gap and demand an explanation for the difference.
• Request human review. Formally request that a qualified human adjuster independently review the claim, not simply confirm the AI output.
• File a complaint with the state insurance department. If the insurer refuses to disclose AI use, refuses human review, or maintains an AI-driven denial without adequate justification, a complaint to the California Department of Insurance (or the applicable state regulator) is appropriate.
• Consult an attorney. AI-driven denials may raise bad faith issues, particularly if the insurer relied on a flawed algorithm, failed to conduct a proper investigation, or refused to allow meaningful human review.

The Bigger Picture: Accountability and Transparency

The core problem with AI in insurance claims is not the technology itself — it is the lack of accountability. When a human adjuster makes a bad decision, there is a person who can be questioned, deposed, and held accountable. When an algorithm makes a bad decision, insurers can hide behind the complexity of the system, claiming the output was objective and data-driven when it was neither.

The regulatory response is still catching up. The NAIC framework and state-level legislation like California’s SB-1120 are important steps, but enforcement remains uneven. Policyholders cannot assume that regulators are monitoring every AI-driven claims decision. The most effective protection remains the same as it has always been: documentation, independent evidence, persistence, and a willingness to escalate when an insurer’s position is unreasonable.

AI will continue to play a growing role in insurance claims. Policyholders who understand how these systems work — and where they fail — are better positioned to identify unfair treatment and fight for the benefits they purchased. An algorithm that says “no” is not the final word. It is the beginning of the conversation.