Part 3 of 4 in a series on AI and claim denials.
Part 1 of this series described the health insurance laws that limit AI-only denials. Part 2 found no comparable P&C statute and pointed to adjuster licensing as a partial substitute. Litigation is the other place where limits are being set. Few cases involve AI in P&C claims directly. The closest P&C case law concerns total-loss valuation software. Read with the health insurance cases, those suits show where a claims organization may encounter future exposure..
Bulk decisioning leveraging a model or AI automation may be challenged
Two health insurance cases make this allegation directly. Plaintiffs in Kisting-Leung v. Cigna allege that Cigna used an algorithm called PxDx to deny claims in bulk without individual physician review. Plaintiffs in Estate of Lokken v. UnitedHealth Group allege that a model called nH Predict replaced clinical judgment in ending Medicare Advantage coverage for post-acute care. These remain allegations pending resolution of the cases.
In P&C, Huskey v. State Farm centers on claims algorithms. Two Black homeowners alleged that a TIer 1 carrier’s automated tools flagged Black policyholders' claims for extra scrutiny more often than white policyholders' claims, which delayed payment. They sued for disparate impact under the Fair Housing Act.
The total-loss suits, including Chadwick v. State Farm and Clippinger v. State Farm, challenge a "typical negotiation adjustment" in valuation reports that lowered comparable-vehicle prices. Some variation of the practice challenged in Chadwick was used by other carriers as well to expedite and partially “automate” the calculation of total loss settlements in auto claims. The practice at its core was intended to provide expedited resolution to customers but it was how the practice was implemented which was challenged.
These complaints focus on allegations of unfair claims practices. The NAIC model act treats refusing to pay a claim without a reasonable investigation as an unfair practice, and it requires a reasonable explanation of the basis for a denial. It reaches conduct frequent enough to indicate a general business practice. A rule built into a tool runs on every claim it touches. If the rule is wrong, the pattern exists from the first day. Thus inappropriate or careless leveraging of AI ability to expedite decisioning could have an impact at scale. This highlights the importance of building with trusted partners who understand the pitfalls and can build to avoid them..
Courts have let contract and discrimination claims proceed
Most of these rulings come at the pleading stage. Denying a motion to dismiss means the allegations, if proven, could support a claim. It is not a finding of liability.
In Lokken (a health related suit), the court in February 2025 dismissed five of seven counts, including bad faith, and let the breach-of-contract and implied-covenant claims proceed. Those claims turn on coverage documents that said clinical staff and physicians would make claim decisions. In Huskey, the court in September 2023 let the disparate-impact claim under § 3604(b) proceed and declined at that stage to treat it as barred by the McCarran-Ferguson Act.
Chadwick, the P&C case involving vehicle valuation, went further. After the court certified an Arkansas class, a jury found for the class on breach of contract in June 2025. The court preliminarily approved a $15.6 million settlement in March 2026 and granted final approval in July 2026.
The contract theory deserves attention in P&C. Policy forms and denial letters describe who handles a claim via a signature at the end of the policy and on any correspondence. If the file cannot show that this person made the decision, the gap supports a contract claim alongside any bad-faith claim. Again, the majority of litigation stems from denial of coverage outside of P&C but until the advent of AI, it was difficult to automate these types of decisions at scale. The advances in AI and the solutions being developed mean that carriers who adopt them should plan ahead.
Courts have also upheld automated valuation
Carriers have won as well. In Signor v. Safeco, the Eleventh Circuit in July 2023 held that a valuation built on a vendor's comparable-vehicle data complied with Florida's total-loss statute, § 626.9743. That valuation lowered each comparable vehicle's price by a uniform condition adjustment, then added $589 because the plaintiff's vehicle was in above-average condition. In Clippinger, the en banc Sixth Circuit in April 2026 voted 10-7 to reverse certification of a Tennessee class of about 90,000 policyholders. The majority held that proving actual cash value would require evidence about each vehicle meaning a class was an unsuitable means of bringing suit as each claimant’s case was unique.
Both defenses rest on individual facts. The facts that defeated class treatment in Clippinger are the ones an adjuster needs in each file to support the payment. In Signor the valuation moved value up as well as down, and a tool that can do both may be easier to defend than one that only discounts.
Claims-handling statutes apply claim by claim
Oubre v. Louisiana Citizens Fair Plan is a delay case with no AI or automated decision in it, only a form letter sent in bulk. We cite it for two points. Louisiana law required insurers to begin loss adjustment within thirty days of notice of a catastrophic loss. In a 4-3 decision in December 2011, the Louisiana Supreme Court held that the penalty required only notice and inaction beyond that period, with no showing of bad faith. It reinstated a class judgment of $92,865,000, or $5,000 per claim. Where a statute sets a fixed claim-handling duty, a missed step is penalized claim by claim.
The second point is what counts as beginning the adjustment. After Hurricanes Katrina and Rita, Citizens pre-printed about 70,000 advance checks of $1,500 and issued them with a form letter stating that it would investigate the claim later. Citizens argued that the checks followed an en masse evaluation using flyovers and aerial surveillance. The court held that the advances did not initiate loss adjustment, which requires "a substantive and affirmative step to accumulate the facts necessary to evaluate the actual underlying claim." The checks went out on each policyholder's word, "without any step taken by Citizens to evaluate the underlying claims." What counted was an adjuster contacting the insured to set an inspection or inspecting the property. An automated step that misses such a duty misses it on every claim routed through it. A bulk communication does not substitute for action on the individual claim. This is particularly relevant where AI is used to create similar bulk actions.
How the tools were built is now a discovery question
We found no court that has held the use of AI in a claim decision unlawful in itself. The open questions concern proof. In March 2026, a magistrate judge in Lokken ordered UnitedHealth to produce records on how nH Predict was developed and used, though not the data, rules, source code, or medical guidelines it is based on. As of September 2026, Lokken and Huskey were both in discovery.
Plaintiffs will ask what a tool was designed to do and what the adjuster did with its output. The answer sits in the claim file and in the AI governance program that the NAIC model bulletin, covered in Part 2, asks insurers to maintain. Part 4 covers what that record should show.
Sources: Kisting-Leung v. Cigna Corp., No. 2:23-cv-01477 (E.D. Cal.) (order granting in part and denying in part motion to dismiss, Mar. 31, 2025). Estate of Gene B. Lokken v. UnitedHealth Group, Inc., No. 0:23-cv-03514 (D. Minn.) (complaint, Nov. 14, 2023; order on motion to dismiss, Feb. 13, 2025; order on motion to compel, Mar. 9, 2026). Huskey v. State Farm Fire & Cas. Co., No. 1:22-cv-07014 (N.D. Ill.) (amended complaint; order on motion to dismiss, Sept. 11, 2023). Chadwick v. State Farm Mut. Auto. Ins. Co., No. 4:21-cv-01161 (E.D. Ark.) (class certification order, Mar. 18, 2024; jury verdict, June 2025; order preliminarily approving settlement, Mar. 27, 2026; order approving settlement and judgment, July 16, 2026). Clippinger v. State Farm Auto. Ins. Co., No. 24-5421 (6th Cir. Apr. 24, 2026) (en banc). Signor v. Safeco Ins. Co. of Ill., 72 F.4th 1223 (11th Cir. 2023), and Fla. Stat. § 626.9743. Oubre v. Louisiana Citizens Fair Plan, No. 2011-C-0097 (La. Dec. 16, 2011), 79 So. 3d 987, 1004-06; La. R.S. 22:658(A)(3) and 22:1220 (renumbered in 2009 as La. R.S. 22:1892 and 22:1973; 22:1973 was repealed effective July 1, 2024, by S.B. 323 (2024 Reg. Sess.), and the good-faith duty now appears in La. R.S. 22:1892(I)). NAIC Unfair Claims Settlement Practices Act (Model 900), Sections 3 and 4. NAIC Model Bulletin: Use of Artificial Intelligence Systems by Insurers (adopted December 4, 2023). Case status is as of September 2026.
This is Part 3 of 4. Part 4 covers what "sole basis" has to mean in practice.
