Lawmakers debate bill to curb insurer use of AI that overrides clinicians
Loading...
Summary
Sponsor Rep. Grama proposed HB 14-06 to prohibit insurers' deployment of AI systems that change or override clinicians' coding and clinical determinations, require documentation of AI use for audits, and treat violations as unfair insurance practices; technical witnesses warned of opaque 'black-box' models while insurers and the Insurance Department pointed to existing human-review safeguards and ongoing rulemaking.
Representative Grama told the committee HB 14-06 would protect clinicians' professional judgment by preventing insurers from using AI systems to change provider codes or otherwise override clinical coding decisions without a qualified human review.
"Clinical coding decisions should remain with the clinicians who actually examine and treat patients," the sponsor said, describing instances elsewhere where automated processes were used to deny claims and citing fast automated processing times and high overturn rates in outside litigation.
Technical witnesses backed the sponsor's concerns. Andrew Horn, a software engineer who built systems for a large insurer, told lawmakers that modern AI models are opaque and probabilistic and that allowing them to make final coverage decisions risks unpredictable and unexplainable denials. "These AI systems are sophisticated random number generators... They don't understand medicine," Horn said.
Patients who testified described cases they said were harmed when claims were denied after automated review. Sharon Bossi told the committee that an insurer initially denied her surgeon's recommendation; by the time she obtained other coverage, a lump had grown and required larger surgery.
Insurers and the Insurance Department urged caution. Michelle Heaton said the department has issued guidance and is drafting rules on AI; she told the committee that under current DOI rules a denial must be issued with an explanation and that denials must be reviewed by a person. Anthem's Sabrina Dunlap said her company does not use AI to make final denial decisions and warned that overly broad language could interfere with accepted, beneficial uses of AI for fraud detection and workflow.
Committee members and witnesses agreed the question of human accountability and audit trails is important; several members asked DOI and stakeholders to refine definitions and reporting mechanics. The committee did not record a final vote; the sponsor and multiple witnesses asked for additional drafting to produce enforceable language and implementation guidance for the Insurance Department.
