A study published in the health policy journal Health Affairs on October 5 examined the consequences of artificial intelligence’s growing role in health insurance claim denial appeals, adding academic scrutiny to a trend that has drawn increasing public attention: insurers using AI not just to process and sometimes deny claims, but also shaping the appeals process patients use to challenge those denials. The research, authored by a team including Lo, Long, Wallace, Salaga and Pestaina, arrives as AI’s role in insurance decision-making has become one of the more contentious intersections of health technology and patient rights.
A two-sided automation trend
The dynamic researchers are examining cuts both ways: insurers have increasingly deployed AI systems to process and sometimes deny claims at scale, while separately, new AI-powered tools have emerged to help patients and providers draft more effective appeals challenging those denials — raising the possibility of an automated arms race where AI systems on both sides of a claims dispute negotiate with each other, with uncertain consequences for the humans whose care is actually at stake.
The scale of AI in claims processing
Major insurers have invested heavily in AI claims infrastructure — UnitedHealth has committed roughly $3 billion toward rebuilding its claims processing systems around AI, with its Optum Real platform reportedly processing 500 million claims in a single quarter and projecting 2.5 billion transactions by year-end. Aetna launched a second-generation AI claims tool in 2026 that the company says cut processing time by more than 20 percent, illustrating the scale at which AI has already been integrated into the core claims workflow across the insurance industry.
The patient risk the research flags
Patient advocates and some researchers have raised concerns that AI-driven claims denial systems can deny coverage at a pace and scale that overwhelms patients’ ability to challenge decisions through traditional appeals processes, particularly for patients who lack the health literacy or time to navigate complex appeal procedures even with AI assistance. The emergence of AI-powered tools to help patients fight back against denials represents a partial response to this concern, though it also raises questions about whether AI-versus-AI dispute resolution produces outcomes that are actually fairer, or simply faster and more opaque.
The regulatory response so far
At least one state, Connecticut, has moved to restrict the practice directly, banning AI-only health insurance denials for state employees as of September — a narrow but notable regulatory intervention that other states may look to as a model. The move reflects growing bipartisan unease among state lawmakers about allowing coverage decisions affecting patient health and finances to be made entirely by automated systems without meaningful human review.
The insurers’ defense
Insurance industry representatives have argued that AI claims processing improves consistency and speed for the vast majority of straightforward claims, freeing human reviewers to focus attention on the more complex or ambiguous cases where human judgment adds the most value — a framing that positions AI as a triage tool rather than a replacement for human decision-making in contested cases. Insurers have also pointed to publicly reported processing time improvements, like Aetna’s 20 percent figure, as evidence that AI adoption is delivering genuine efficiency gains rather than simply cutting costs by denying more claims.
What it means next
With the Health Affairs study adding to a growing body of academic scrutiny on AI’s role in insurance decision-making, the debate over how much human oversight AI-assisted claims and appeals systems require is likely to intensify at both the state and federal level. Connecticut’s narrow ban on AI-only denials for state employees may prove to be an early test case other states watch closely before deciding whether to pursue similar or more expansive restrictions on automated coverage decisions.
Photo: RDNE Stock project / PEXELS via Pexels
