Minnesota Bars Automation Alone in Prior Authorization Denials

Minnesota has become the latest state to restrict how health insurers use artificial intelligence in prior authorization decisions. A bill passed by wide margins in the state House on May 16 and in the Senate on May 17 prohibits a utilization review organization from denying a preauthorization request on the strength of automated processing alone. Governor Tim Walz signed it on May 27, 2026 as part of the commerce omnibus bill, now Laws of Minnesota 2026, chapter 124.

Effective January 1, 2027, the law reaches health insurance plans offered, sold, issued, or renewed on or after that date. The enacted language bars a utilization review organization from using any form of automated processing alone, without a clinician review by an appropriate health professional, when it makes an adverse determination. Two things follow from how that is worded. The statute never says artificial intelligence, so an older rules engine falls under it as squarely as a model does. And the reviewer is a physician only where the rest of section 62M.09 already requires one, meaning adverse determinations made for clinical reasons; a prescription-drug denial may be reviewed by a licensed pharmacist instead.

Four states got there first, and the variations between them matter more than the shared direction. Maryland’s Chapter 747 adds a line to the quarterly adverse-decision report carriers already file, asking whether an artificial intelligence, algorithm, or other software tool was used, and separately requires that such a tool not replace the role of a health care provider in the determination. Nebraska’s LB 77 bars an AI-based algorithm from being the sole basis of a denial, requires disclosure of AI use to the department, to network providers, to enrollees, and on a public website, and lets the department audit the automated system at any time. Texas SB 815 goes furthest on the prohibition itself: an automated decision system may not make an adverse determination wholly or partly, and the commissioner may audit that use at any time. Arizona’s HB 2175 never mentions AI. It requires the medical director to individually review a medical-necessity denial and forbids relying solely on recommendations from any other source, which catches a model without naming one. Illinois is often counted in this group; its bill passed the House and then stalled in the Senate in June 2025, and it is not law.

From January 1, 2027, every automated denial issued under a covered Minnesota plan needs a clinician’s review before it goes out, which is a workflow change more than a policy one. That will likely require changes to documentation standards, escalation rules, and training for clinical review staff. The bill also creates a compliance benchmark that other states may copy, particularly as litigation over AI-driven denials continues to attract attention. UnitedHealth Group, Cigna, and Humana have all faced allegations that they used AI algorithms to deny claims.

The Minnesota approach puts its weight in a different place. Texas and Nebraska both hand their regulators a standing right to audit the automated system itself. Maryland runs its signal through a report carriers already file. Minnesota makes the clinician review the central control, which is easier to evidence but harder to scale. Carriers that operate across multiple states should not assume Minnesota’s standard will stay unique; the pattern points toward a patchwork of human-review requirements spreading through 2026 and 2027.

These states converge on one bar: automation alone cannot carry a denial. They diverge on how it is enforced, and that is the part that shapes the workflow. Minnesota audits through the clinician-review record, Texas and Nebraska through direct audit authority over the system, Maryland through a line item in a filing that already exists. One workflow can satisfy all three, but only if it produces evidence rather than just a decision, which is the design question sitting under every state AI rule in health insurance.