Insurance companies are quietly folding artificial intelligence into the claim decisions that determine whether people keep their disability benefits, their life insurance payouts, and their financial futures. No insurer has ever told a claimant outright that a machine wrote their denial letter. But after three decades litigating against the country’s largest carriers, I am seeing the signs everywhere, and the legal system is only beginning to catch up.
What a Denial Looks Like When AI Wrote It
For years, a denial letter meant a claims examiner, however biased or mistaken, had read the file. That assumption is eroding.
Just this fall, we worked an appeal built around a neuropsychological evaluation, the gold standard for documenting cognitive impairment because it rests on standardized, objective testing rather than a patient’s own account of their symptoms. The insurer’s hired consultant came back saying there was no evidence of cognitive impairment, and never mentioned the evaluation at all. When we pushed back, the file went to the same consultant a second time. His answer: neuropsychological testing is “just another data set.” It is not. It is objective proof, the same way an X-ray proves a fracture.
We are seeing this pattern across files: detailed legal citations submitted in an appeal, met with a response that shows no engagement with what was argued. Letters that read as if a machine assembled them from a checklist rather than a person who took responsibility for understanding the claim.
Discovery Into AI Systems: A New Litigation Tool
No insurer is going to admit, unprompted, that an algorithm wrote a denial. That is exactly why discovery now has to reach further than the claim file itself.
Under ERISA, courts have long recognized that a conflict of interest, an insurer that both decides and pays claims, is a factor courts weigh when reviewing a denial (Metropolitan Life Insurance Co. v. Glenn, 2008). An insurer that cannot explain how it reached a decision, because the decision ran through a system no one at the company fully understands, presents a procedural irregularity of the same kind. That opens the door to discovery aimed at the AI system itself: what model was used, what data trained it, what instructions shaped its output, who at the company reviewed the result before it went out, and what the vendor contract says about human oversight.
This is new ground, and insurers will resist it. But claimants’ counsel who do not ask these questions are leaving a central fact about how the denial was made undiscovered.
Fiduciary Breach When No Human Reviewed the Claim
ERISA’s claims regulations require a “full and fair review” of every denied claim. That standard was written with a person in mind, someone who reads the file, weighs the evidence, and takes responsibility for the conclusion.
A denial generated primarily by a machine, with no meaningful human judgment behind it, undercuts an insurer’s claim to have exercised the discretion its own policy grants it. That argument supports pushing for the more exacting standard of review courts apply when a plan administrator has not done what the policy and the regulations require. It also strengthens broader bad faith claims in life and AD&D cases governed by state law, where an insurer’s duty to investigate a claim fairly has never included outsourcing that duty to software no one at the company can explain.
We are building this argument into our appeals and our litigation strategy now, not waiting for a court to catch up to the technology first.
Why the NAIC’s New Evaluation Tool Matters for Claimants
Regulators are watching this too. The National Association of Insurance Commissioners adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers in December 2023, and nearly half the states have adopted it since, each requiring insurers to maintain a written AI governance program covering risk management, testing, vendor oversight, and documentation.
In 2026, the NAIC began piloting an AI Systems Evaluation Tool across a dozen states. The tool asks insurers to disclose which systems merely support a person’s decision, which augment it, and which automate it outright, with particular attention to AI used in claims handling. That distinction matters to claimants, because an insurer that has automated a decision has a much harder time explaining why no one engaged with the specific evidence submitted.
For claimants, the real value of this regulatory record is what it generates: governance documents, system inventories, and vendor disclosures that did not exist before and that become relevant evidence once a claim is denied. We intend to use them.
How We Are Preparing for This Fight
We use AI ourselves, carefully. It is a genuine time saver for working through claim files that run to thousands of pages, and it can help pressure-test an argument or sharpen how we present it. But we learned quickly that an AI vendor’s file review can miss hundreds of pages. A missed page in a disability claim is not a rounding error. It is someone’s benefits. We do not use consumer AI platforms for client matters at all, because they are not confidential, and our clients’ medical and financial information does not belong in a system we do not control.
What we have changed is how we build a record. The more thoroughly we document a claim, with every relevant citation and piece of medical evidence placed directly in the file, the harder it becomes for any reviewer, human or algorithmic, to overlook what matters. A federal appellate judge once observed that judges are not expected to go looking for the argument a party failed to make. The same logic applies with even more force when the reviewer on the other side may be software: we are not leaving anything for it to miss.
Related Article: The ERISA “Administrative” Record: How to Build the File That Wins Your Disability AppealWhat to Do if Your Claim Was Denied
If a denial letter ignores evidence you know was submitted, misstates your diagnosis, or reads like no one engaged with your file, say so, immediately, to a lawyer. Those discrepancies are not just frustrating. They may be the clearest evidence of how the decision was made.
The clock does not wait for the technology to sort itself out. Long-term disability claims generally carry a 180-day appeal deadline, and some health benefit appeals must be filed within days. Reach out the day you receive a denial, not after you decide whether you have the energy to fight it. We will tell you quickly whether we can help, and if AI played a role in getting your claim wrong, we intend to make that part of the case.
