Lawmakers weigh model 'AI Medical Services Act' to license clinical AI, labor commissioner warns on costs and authority

Members of the Montana Legislature’s Economic Affairs Interim Committee heard detailed testimony on a model “AI Medical Services Act” that would license clinical artificial intelligence as a type of provider and allow provisional sandbox deployments ahead of full licensure.

Adam Meyer, director of health care policy at the Cicero Institute, told the committee the proposal focuses on regulating the clinical service performed by AI rather than the underlying software. Under the model, AI services would receive a two‑year provisional license during a supervised pilot period. ‘‘If you meet or exceed any human benchmarks for that particular service line, after that two‑year period you can be granted a full license, which would allow them to practice statewide,’’ Meyer said.

Meyer described a risk‑tiered framework that combines an AI autonomy level (informational, advisory, supervised autonomous, fully autonomous) with the clinical severity of conditions (preventive, chronic/noncritical, critical/time‑sensitive). Licenses would carry disclosure and auditability requirements, human‑escalation pathways, malpractice bonding and continuous performance monitoring. The draft also proposes a payment pathway by asking the Department of Insurance to provide guidance on reimbursement and a state MPI number until a federal identifier exists.

Why it matters: proponents say licensing clinical AI could expand access—particularly in rural areas facing provider shortages—while preserving patient protections and avenues for liability and oversight. Meyer argued the choice is not AI versus no AI, but unregulated consumer apps versus regulated clinical services integrated into the health system.

Swanson: funding, technical capacity, authority

Sarah Swanson, Commissioner of the Department of Labor and Industry, said the department is interested but has major concerns. ‘‘I think $500,000 for a brand new program start up might be a little bit on the low side,’’ Swanson said, referencing cost estimates in the draft. She told lawmakers the department currently lacks staff with the necessary technical expertise to test models, detect model drift and manage continuous monitoring. Swanson urged the committee to identify a sustainable funding source rather than shift costs onto existing licensees through higher fees.

Swanson also flagged a legal and operational gap if complaints concern non‑human systems: current disciplinary processes under Title 37 apply to licensees. ‘‘If an AI tool begins to drift and has caused or has a risk of harming people, what do we do? Who has the kill switch?’’ she asked. She recommended keeping boards that already oversee practice acts involved and using designees from those boards on any new advisory or oversight body.

Committee questions and follow‑up

Committee members asked about data sharing, patient consent and whether AI could ever be permitted to prescribe scheduled narcotics. Meyer said the model excludes prescribing of scheduled narcotics and that most early use cases would remain supervised‑autonomous with human oversight. He pointed to existing state sandboxes (for example Utah) and emphasized the draft’s pathway out of a provisional license into full licensure.

Next steps

Members did not vote on legislation at the hearing. Staff and interested members signaled they will continue refining the draft, and agencies noted that any costs and jurisdictional changes would be vetted through the fiscal note process and additional stakeholder consultation.

Sources: Testimony from Adam Meyer (Cicero Institute) and Sarah Swanson (Commissioner, Montana Department of Labor and Industry) before the Economic Affairs Interim Committee.