Artificial Intelligence (“AI”) was once a thing of the future. In the employee benefits world, the future is now: plan sponsors, administrators and service providers are using AI in the benefit claim determination process. This raises a question: how can AI be utilized under the Employee Retirement Income Security Act of 1974 (ERISA) consistent with the terms of the plan? Thus far, one notable lawsuit addressing this issue alleges that a claims administrator violated ERISA’s fiduciary duty by using an AI-based algorithm to deny benefit claims when plan terms stated claims would be reviewed by a medical director.
This article discusses the use of AI in determining benefits claims and how one pending court case is impacting its viability. In addition, we address how California is attempting to shape the path forward for AI.
Pending Benefit Claim Case in Which Use of AI Is at Issue
In Kisting-Leung v. Cigna Corporation, et al., No. 2:23-cv-01477 (E.D. Cal. July 24, 2023) participants in different Cigna-sponsored health plans filed a class-action lawsuit. The plaintiffs’ sued the plans’ common claims administrator, claiming its use of an automated algorithm to process benefit claims was contrary to plan terms, which required a medical director to process benefit claims. Plaintiffs alleged that Cigna’s AI tool, called “PxDx,” wrongfully enabled Cigna to circumvent the plan’s requirement that a medical director must review and assess medical claims, any prior medical or surgical history, and treatment.
The plaintiffs’ first two complaints only included state law causes of action, such as breach of the implied covenant of good faith and fair dealing, violation of California unfair competition law (§17200), and intentional interference with contractual relations. It was not until the third and final amended complaint that the plaintiffs replaced all but one of their state law causes of action (violation of California unfair competition law) with ERISA claims: one under ERISA §502(a)(1)(B) for wrongful denial of benefits and the other for breach of fiduciary duty under ERISA §502(a)(3).
Cigna moved to dismiss the third amended complaint, arguing that three of the six named plaintiffs’ lacked standing because Cigna did not deny treatment for these claims using PxDx, which was corroborated by a declaration from Cigna’s medical officer. Cigna also argued that the plaintiffs’ claim to enforce plan rights under §502(a)(1)(B) failed because the plaintiffs did not allege any plan provisions that were breached and did not show exhaustion of their administrative remedies, as required by ERISA §503.
As for the breach of fiduciary duty claim under ERISA §502(a)(3), Cigna argued that this claim failed because “Cigna’s use of PxDx to deny plaintiffs’ claims violated neither the terms of plaintiffs’ plans nor ERISA,” and as a result, plaintiffs’ did not allege a breach of fiduciary duty. Cigna further argued that plaintiffs’ did not state a claim for equitable relief, and the claim is entirely duplicative of the §502(a)(1)(B) claim. Cigna did not dispute that it was an ERISA fiduciary acting in its respective fiduciary capacities but only contested whether Cigna violated ERISA-imposed fiduciary obligations. Furthermore, in regard to whether Cigna violated its fiduciary duty, it argued that because doctors “are the ones using the algorithm to ‘automatically deny payments in batches of hundreds or thousands at a time for treatments that do not match certain pre-set criteria,’ defendants are in compliance with the plan requirement that medical necessity decisions be made by a Medical Director.” Additionally, Cigna argued that using PxDx and its algorithms to process claims was within its discretionary authority to interpret plan terms, noting that it had issued disclosures to doctors and participants regarding its use of PxDx.
On March 30, 2025, the court granted in part and denied in part Cigna’s motion to dismiss. In response to whether the three named plaintiffs had standing, the court stated that all plaintiffs had standing to bring claims under ERISA §502(a)(1)(B), for wrongful denial of benefits. However, the court ruled that the plaintiffs had failed to identify the specific plan terms that made them eligible for benefits. As a result, the plaintiffs had failed to sufficiently allege facts stating a claim under §502(a)(1)(B) and, thus, the benefit denial claims failed. As to the breach of fiduciary duty claim under ERISA §502(a)(3), the court stated that the plaintiffs lacked standing to bring their claims because the complaint failed to state facts sufficient to plausibly allege that PxDx was used to deny their claims. More specifically, because the plaintiffs’ claim determinations “were not made using the PxDx algorithm, denial of their claims is not fairly traceable to the defendants’ challenged behavior.”
The court denied the motion to dismiss as to the fiduciary breach claims on behalf of the three remaining plaintiffs. The court disagreed with Cigna’s argument that because a doctor oversaw PxDx, it complied with plan terms. The court noted that, “the standard used in evaluating an ERISA plan administrator’s interpretation of a plan term varies depending, in part, on whether the plan gives the plan administrator discretion to interpret the terms of the plan.” More specifically, the court stated that “even assuming without deciding that this deferential standard applies here,” the “defendants’ interpretation of the plan provision requiring determinations of medical necessity be made by a medical director—as allowing an algorithm to make the decision so long as a medical director pushes the button—conflicts with the plain language of the plan and constitutes an abuse of discretion.” As a result, the court concluded that the plaintiffs adequately alleged a violation of plan terms and allowed the breach of fiduciary duty claim to proceed.
The court further ordered that the injunctive relief sought by the plaintiffs was an appropriate equitable relief available under ERISA §502(a)(3). More specifically, the court agreed with plaintiffs that if there was a violation of plan terms, the injunctive relief enjoining Cigna from “continuing its improper and unlawful claim handling practices as set forth” was an available and valid relief.
As of April 11, 2025, the plaintiffs’ filed a notice of intent not to amend their complaint. We will continue to monitor the progress of this case, as the outcome could have significant implications for the use of AI in the administration of health care claims.
California’s Legislative Action
To date, Congress has not enacted any legislation that would preclude (or support) the use of AI for benefit determinations in ERISA-covered plans. At the state level, several states have enacted laws to regulate AI use in health plans. As an example, effective January 1, 2025, California passed SB 1120, which modifies §1367.01 of California’s Health and Safety Code. SB 1120 provides that AI tools cannot replace healthcare decision-making and autonomously deny, delay, or modify care. It further provides that final determinations regarding medical necessity must be made by licensed physicians or health care professionals competent to evaluate the clinical issues. Under SB 1120, California is ensuring that a qualified human is involved in the review and determination process.
In general, state laws governing insurance – such as §1367.01 of California’s Health and Safety Code – apply to fully insured ERISA plans but not to self-funded plans. The administration of self-funded health plans is governed exclusively by ERISA. Therefore, state laws that regulate the use of AI by health plans likely apply to fully insured ERISA plans but not to self-funded plans, due to ERISA’s preemption of state laws.
Conclusion
Kisting-Leung may be one of the first of a number of lawsuits challenging the proper role of AI in the benefit claim administration process. This highlights the need for health plan sponsors to carefully review their plan documents and assess whether, in operation, their claims administration processes follow plan terms. It may be that AI can play a valuable role in the benefit plan claims administration; however, as highlighted by Kisting-Leung, it can do so only if properly provided for under plan terms.
