Your chronic-condition prescription is up for renewal. The system evaluating it isn’t your doctor recalling your last visit; it’s an algorithm working from your chart, and in Utah, that scenario is now authorized under state supervision.
The Utah Office of Artificial Intelligence Policy has approved three government health AI pilots under its regulatory sandbox, covering clinical decisions, medications, and patients enrolled in active care. The oversight architecture is more substantive than most state programs have attempted. The question the program still hasn’t answered publicly is what happens when something goes wrong, and who finds out first.
What Utah Actually Approved
Three companies now have state authorization to automate parts of clinical care, each at a different level of risk.
Nolla Health received authorization to evaluate adults with mild-to-moderate acne and support initial topical prescriptions. Licensed clinicians currently review and approve those prescriptions before they reach a pharmacy.
Utah may eventually allow prescriptions to proceed without that prior review. The conditions that would permit that transition are not publicly specified in the reviewed announcements.
August AI is piloting AI-assisted renewals for existing, non-controlled prescriptions used to manage chronic conditions. The system starts with a formulary of approximately 148 medications, prioritizing lower-risk drug classes first.
The system will not issue new prescriptions, change dosages, or substitute medications. A Utah-licensed physician reviews each renewal before it is authorized.
Expect Fitness will test AI-supported telehealth for pelvic-floor physical therapy, focusing on women’s health including pregnancy and postpartum care. Licensed physical therapists retain oversight of patient intake, treatment planning, and follow-up.
Utah Built Real Infrastructure Around These Pilots
Six independent evaluators, two major health-system partnerships, and 12-month monitored terms give the program structural weight.
Utah did not simply hand companies a permission slip. Six independent evaluators will stress-test guardrails, inspect decision logs, and assess patient outcomes alongside state regulators.
Those evaluators include the Coalition for Health AI, Clarion AI Partners, Glacis Technologies, mpathic AI, Vega Health, and Stanford’s Clinical Excellence Research Center. Master agreements with University of Utah Health and Intermountain Health create a formal governance layer rather than ad hoc approvals.
Pilots run for 12 months, with state-monitored performance requirements before any expansion is considered. This is not a sandbox in name only.
Where Public Accountability Is Still Missing
The oversight framework looks rigorous; the reviewed announcements do not explain how or when Utah will publicly report on results.
The reviewed announcements do not specify how Utah will publish model error rates, adverse events, demographic disparities, patient complaints, or auditor findings. Without that disclosure, patients, clinicians, and independent observers cannot assess performance claims on their own.
The Nolla pilot warrants particular scrutiny. Moving from clinician-reviewed initial prescriptions to unsupervised AI prescribing is a materially different risk profile, not an incremental adjustment. The criteria that would trigger or prevent that transition should be stated publicly before the pilot ends, not determined afterward.
Third-party auditors are only as independent as their contracts allow. Utah should publish those evaluator agreements, conflict-of-interest disclosures, and the evidentiary standards auditors will apply.
Per a Vega Health statement, evaluators will assess anticipated benefits, potential failure modes, and evidence that could change an approval or rejection decision. Those criteria should be public, not summarized in a press release.
Transparency Is the Only Real Test
If Utah publishes the data, this program becomes a model. If it doesn’t, it becomes a cautionary tale with good branding.
If Utah releases outcome data, error rates, and auditor findings in enough detail for other regulators to learn from, this becomes a credible template for supervised clinical AI. If it doesn’t, it joins a long list of pilots that promised accountability and delivered announcements. The difference between those two outcomes is public disclosure, enforced before the 12-month clock runs out.
