A public disagreement between billionaire investor Mark Cuban (pictured, left) and venture capitalist Marc Andreessen (pictured, right) over whether AI has surpassed human doctors has drawn fresh attention to a less celebrated side of the technology: how health insurers are using AI to manage and in some cases deny claims.
The exchange began on July 12 when Andreessen posted that AI is already a better doctor than 99.99% of human doctors, amplifying similar comments from OpenAI CEO Sam Altman after OpenAI's health research lead shared results from GPT-5.6 showing physicians rated the model's answers as flawless in roughly a fifth to a quarter of blinded cases against about 10% for physician-written responses. Cuban responded with a flat rejection, arguing that a smarter diagnostic tool does nothing to address the friction patients face once a diagnosis reaches an insurer - and warning that the same companies building better diagnostic AI are also building AI systems designed to slow down or reject claims.
Cuban's warning is not speculative. The NAIC's most recent Artificial Intelligence and Machine Learning survey, drawing on responses from 93 companies across 16 states, found 84% of health insurers currently use AI or machine learning in some capacity, with 12% specifically applying it to denying prior authorizations. Nearly a third of health insurers do not regularly test their models for bias or discrimination despite NAIC guidance recommending it since December 2023. That governance gap has a specific litigation expression: Cigna faces a federal lawsuit - Kisting-Leung v. Cigna - whose complaint cites 2023 ProPublica reporting that its PxDx algorithm denied more than 300,000 claims over a two-month span in 2022 at an average review time of roughly 1.2 seconds per claim. Cigna has called the lawsuit baseless, saying PxDx is not AI-powered and applies to only a small subset of claims. A federal judge allowed the case to proceed in March 2025. UnitedHealthcare and Humana face similar proceedings.
KFF's analysis of federal marketplace data found that fewer than 1% of denied Affordable Care Act claims were appealed in 2023, and insurers upheld 56% of the denials that were challenged - a claims environment in which AI-enabled denial at scale generates limited immediate pushback but accumulating litigation risk. A West Health-Gallup survey released in April found one in four US adults has used AI for a health question, but only 4% of recent users said they strongly trust its accuracy.
California's Physicians Make Decisions Act, in effect since January 2025, bars health insurers from denying medically necessary care based solely on an algorithm and requires licensed human review of such denials. Nearly 30 states have adopted some version of the NAIC's model bulletin on AI governance, and the NAIC's AI Systems Evaluation Tool has been piloted in 12 states since March 2026 with nationwide rollout targeted for later this year.
That state-level push has drawn direct pushback at the federal level. In a statement issued after the White House signed its December 2025 executive order on AI, the NAIC said the order introduces legal uncertainty that could weaken the insurance market by delaying business decisions and postponing consumer protections, calling on the administration to affirm state authority over AI in insurance. The NAIC also pointed to prior letters it sent Congress in December and June on the same issue.
For health insurers, MGAs and brokers, the Cuban-Andreessen exchange is less a tech story than a reminder that AI governance compliance is a live operational requirement. With nearly a third of health insurers still not regularly bias-testing their models, an NAIC-federal standoff over who regulates AI in insurance, and litigation already underway against the sector's largest carriers, insurers deploying AI in utilization review have a narrowing window to document governance processes before the NAIC's evaluation tool reaches full national rollout.
