Payers are denying with AI

A growing number of states now require a clinician to sign off before a payer's algorithm can deny care. More are moving that way. Most states still have nothing on the books.

In April 2026, KFF (the Kaiser Family Foundation) published a tracker of state and federal laws governing the use of artificial intelligence (AI) in prior authorization (PA) and claims review. The picture: California, Texas, Illinois, Maryland, Utah, Washington, and Alabama have enacted laws with teeth. Several more states have bills moving in 2026. The rest have nothing specific on the books.[2]

The backdrop is a National Association of Insurance Commissioners (NAIC) survey released in May 2025. Of 93 large health insurers across 16 states, 84% reported using AI or machine learning (ML) somewhere in their operations. Thirty-seven percent said they use AI specifically for PA today. Fifty-six percent use it for utilization management (UM).[1] None of that was illegal anywhere when the survey ran. The question is where law has since caught up.

California moved first. Under SB 1120, signed September 28, 2024 and effective January 1, 2025, any PA decision that denies or modifies a treatment request must be reviewed by a licensed physician or qualified clinician. An algorithm cannot be the last word. Picture a 20-clinician psychiatry group in Los Angeles: early 2025, two of their commercial payers were returning same-day PA decisions on medication management requests, faster than any human reviewer could have read the submitted records. Under SB 1120, that process is now presumptively illegal. Their billing director filed with the California Department of Managed Health Care; one payer revised its workflow.[2]

Federal law is not filling the gap in the other 43 states. The One Big Beautiful Bill Act (OBBBA), signed July 4, 2025, bars the Centers for Medicare and Medicaid Services (CMS) from regulating Medicare Advantage AI-based coverage decisions through administrative rulemaking. Federal AI-denial legislation has stalled. The NAIC Model Bulletin, adopted by 24+ states in 2025, creates governance expectations but not binding law.

The exposure is not that AI is reviewing claims. It is speed. AI returns a denial before a clinician can build the documentation that would reverse it. The concrete move: find out which of your top three payers use AI screening for PA, and whether you are in a state that requires human review before denial. If you are in California, Washington, or Illinois, you have leverage to push back on automated denials. If you are not, your note has to do the arguing up front, before the algorithm ever sees it.