Artificial Intelligence (AI) is constantly revolutionizing industries, and claims handling is no exception. These advancements come with growing pains: the use of AI in claims determinations has become the subject of bad-faith litigation. Although AI is relatively new, the bad faith arguments are not. Bad faith lawsuits concerning the use of technology in claims handling date back over 40 years. See Blue Cross and Blue Shield of Alabama v. Granger, 461 So. 2d 1320, 1327 (1984) (noting that “[a]n insurer which, without more, seeks to serve its insureds by computerizing its claims system should not be penalized for its efforts when, through no intentional act of the insurer, a claim runs afoul of the system.”)
Recently, United Healthcare was sued for its use of AI in claims handling. See First Amended Complaint, Est. of Lokken v. UnitedHealth Grp., Inc., No. 0:23-cv-03514-JRT-DTS, 2024 WL 2853368, (S.D. Minn. April 5, 2024) (asserting, among other claims, breach of contract, breach of implied covenant of good faith and fair dealing, and bad faith arising out of United Healthcare’s alleged usage of an AI model in Medicare Advantage policy claims determinations, despite knowledge that the model has a 90% error rate).
Although in Estate of Lokken, the court determined plaintiffs’ bad faith claim was due to be dismissed because a bad faith analysis would require the Court to determine whether each denial was reasonable, plaintiffs were permitted to proceed with their breach of contract and breach of the implied covenant of good faith and fair dealing claims. Est. of Lokken v. UnitedHealth Grp., Inc., 766 F. Supp. 3d 835, 848 (D. Minn. 2025). These causes of action survived because plaintiffs alleged United Healthcare represented claims determinations were made by physicians and clinical services staff when it allegedly used AI to make these determinations. Id. To evaluate these claims, the Court would only have to determine whether United Healthcare complied with its own written documents, as opposed to evaluating the standards promulgated by the Medicare Act to determine if each denial was reasonable. Id.
As the law concerning the use of AI continues to develop, how courts have dealt with bad faith suits related to the use of other technology in claims determinations can provide guidance in how courts may consider these issues in the future. For example, in Milhone v. Allstate Ins. Co., 289 F. Supp. 2d 1089 (D. Ariz. 2003), a plaintiff was injured in a car accident and later made a demand to his insurer for $18,600. The insurer used a computer program, COLOSSUS, to determine the value of the claim. The insurer’s first offer to the plaintiff was within the recommended COLOSSUS range. The plaintiff rejected the offer, and the insurer made another offer, above the range recommended by COLOSSUS. After that offer was rejected, the parties proceeded to arbitration, where the plaintiff was awarded $9,000. The plaintiff then filed a bad faith suit, in part because of the insurer’s use of COLOSSUS in its claims handling.
The court granted the insurer’s motion for summary judgment on bad faith. The court noted that the plaintiff failed to argue how the insurer’s usage of a uniform claims evaluation resulted in bad faith. Milhone, 289 F. Supp. at 1101. The court determined that “Plaintiff was made an offer that was above the COLOSSUS range; therefore, the adjusters could overcome any ‘bad faith’ in the COLOSSUS system via a manual adjustment of the amount offered.” Id. Thus, the use of a computer program was not enough to sufficiently allege bad faith.
Similarly, property insurers have been successful at the summary judgment stage on bad faith claims so long as they can prove they do not lack a reasonable basis for using a computer program when determining the value of replacement costs. Sands v. State Farm, No. 5:17-cv-4160, 2018 WL 1693387, at *5 (E.D. Pa. Apr. 6, 2018) (noting that the insurer could rely on Xactimate because it is standard in the industry to estimate replacement costs). However, a carrier has been denied summary judgment as to whether it acted in bad faith or was negligent in allegedly “rely[ing] solely on its computer system to determine … limits that current estimates of the cost of rebuilding suggest to be inadequate.” Lewis v. Allstate Ins. Co., No. 3:15-cv-8074-HRH, 2016 WL 5408332, at *6 (D. Ariz. Sept. 28, 2016).
The overarching theme in past litigation, and the warning that all AI developers give, is the importance of human oversight. As AI becomes more prevalent, it is imperative that adjusters and analysts retain the ability to deviate from AI’s recommendations.