Public support for human review rises as AI denial rules spread
Eighty-two percent of Americans would support a law requiring a human to review any medical claim denied by an artificial intelligence tool. The number lands at the center of a debate that has moved far beyond abstraction: insurers are increasingly using automated systems to sort coverage decisions, and patients are pushing back against a process many view as opaque and too easily disconnected from bedside judgment.
The finding came from an eHealth open enrollment survey conducted Oct. 7-9 through SurveyMonkey among more than 1,000 U.S. adults, with a margin of error of plus or minus 3 percentage points. eHealth is an online health insurance marketplace and licensed agency that says it uses AI to help consumers compare plans. The survey did not say whether any respondents had experienced an AI-driven denial firsthand, but the response pattern suggests broad unease with delegating coverage decisions to software alone.
A single data point captures the political temperature around this issue: human review has become the default demand.
Prior authorization sits at the center of the dispute
The concern is not theoretical. Seven states passed laws this year limiting how health insurers use AI in coverage decisions, and a bipartisan House bill introduced Sept. 1 would require a licensed health care professional to make any private insurance denial that turns on clinical judgment. Those state laws already took effect in Washington, Iowa and Alabama this year, with Colorado, Georgia and Utah scheduled to follow on Jan. 1, 2027. Illinois has a separate law requiring a person to make or review every downcoding decision starting Jan. 1, 2028.
Earlier statutes in California, Arizona, Texas, Nebraska, Connecticut and Maryland add to a patchwork that is tightening the reins on automated utilization management. Yet the patchwork is incomplete. The state laws do not reach self-insured employer plans, which are governed by federal law, and Medicare Advantage generally follows federal standards. That gap matters because it leaves large segments of the insured population outside the reach of the newest state-level protections.
Traditional Medicare has also stepped deeper into prior authorization. On Jan. 1, it began requiring prior authorization for a set list of services in Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington under the Centers for Medicare & Medicaid Services Wasteful and Inappropriate Service Reduction model. Technology companies review those requests using AI and other tools, and they are paid a share of the spending they avert. A clinician employed by the company must review every request the company declines to approve.
Denial volume and overturned decisions fuel skepticism
The practical consequences of automated review are already visible in the federal pilot. Two companies denied 5,944 prior authorization requests in the program’s first three months, based on roughly 1,000 pages of CMS records obtained through a Freedom of Information Act lawsuit and released in September. Virtix, the Washington vendor, denied 53% of the more than 6,000 decisions it made through the end of March, according to a status report sent to CMS. A contractor in Arizona made 14,300 decisions over the same period and denied fewer than 20%.
Virtix has said its approval rate has since risen to nearly 70%. CMS ordered Virtix to file a corrective action plan in June after the company missed required turnaround times. Those processing delays are not a clerical nuisance. When prior authorization slows, treatment plans stall, hospital discharge planning is disrupted, and clinicians inherit more administrative work at the exact moment care decisions need speed.
A separate federal review adds another layer of concern. Medicare Advantage plans overturned 95% of appealed prior authorization denials for skilled nursing facility admission in a June report from the Department of Health and Human Services Office of Inspector General, based on 19 insurers’ June 2024 decisions. Patients and physicians appealed only 18% of those denials. The OIG concluded that the overturn rate indicates some enrollees were initially denied medically necessary care.
Legislative momentum is shifting toward physician oversight
The Doctors Not AI Act of 2026 would go further than the current state laws. Introduced by Rep. Greg Landsman of Ohio, Rep. Buddy Carter of Georgia, Rep. Kim Schrier, M.D., of Washington and Rep. Tom Barrett of Michigan, the bill would bar an AI system from issuing or dictating any denial involving clinical judgment. A licensed professional with training and experience in the service at issue would have to make the decision after independently evaluating the patient’s circumstances, and insurers would have to disclose when AI was used in claims review.
The bill applies to group health plans and to insurers selling group or individual coverage, but not Medicare or Medicaid. MGMA endorsed it on Oct. 16. In a statement at introduction, Schrier said doctors, not automated systems or AI-powered models, should be in charge of critical health care decisions for their patients.
The language of the bill reflects a basic clinical principle: coverage denials should not be detached from the realities of disease severity, function, comorbidity, and expected response to therapy. Algorithms may standardize patterns, but they do not examine the patient, hear the symptom trajectory, or weigh the downstream harm of delay in an individual case.
Delay, wrong advice, and administrative burden converge
The same survey that captured anger over claim denials also found that 58% of insured adults delayed care because of cost. Two-thirds of respondents who had used AI for medical advice said it was sometimes wrong. Those results do not prove a direct link between AI and delayed treatment, but they do show how easily automation can erode confidence when people already struggle to access care affordably.
The policy argument now extends well beyond insurance operations. If patients expect human review before a claim is denied, and if federal data show overturned denials at very high rates, the burden shifts to payers to justify why an automated first pass should stand without physician oversight. A system that repeatedly gets reversed after appeal is not simply efficient and imperfect; it is structurally vulnerable to error at the point where patients can least afford it.
CMS has said it will release a six-month analysis of the model’s outcomes, including turnaround times and preliminary accuracy data. That readout will matter because the current debate rests on early operational signals, state-by-state legal responses, and a public that already appears wary of AI making coverage decisions in the background.
The remaining question is narrower and more concrete: whether insurers and regulators can preserve utilization control without turning clinically consequential denials into machine-driven defaults.