Skip to content

Menu

Network by SubjectChannelsBlogsHomeAboutContact
AI Legal Journal logo
Subscribe
Search
Close
PublishersBlogsNetwork by SubjectChannels
Subscribe

A Deep Dive into Utah’s Healthcare AI Experiment

By Jordan Cohen on November 16, 2026
Email this postTweet this postLike this postShare this post on LinkedIn

On October 5, 2026, the Utah Department of Commerce announced five new healthcare agreements under its AI regulatory sandbox, including a pilot that allows an AI system to issue initial prescriptions for acne treatment. Together with a prescription renewal pilot approved in October 2025, the agreements show in detail how one state is approaching oversight of AI that takes clinical action, including what the tools may do, how physician review changes over time, and what the state requires before that review is reduced.

Supporters see a controlled way to test whether AI can expand access to routine care. Critics, including Utah’s Medical Licensing Board, have objected that the first pilot was launched without the Board’s input and that renewing a prescription calls for a physician’s clinical judgment. This post summarizes the legal framework, the terms of the agreements, and the federal and state context, and closes with the question healthcare organizations may want to watch most closely as the pilots proceed.

How Utah’s Sandbox Works

Utah’s Artificial Intelligence Policy Act, enacted in 2024 and codified at Utah Code Title 13, Chapter 72, created the Office of Artificial Intelligence Policy (OAIP) within the Department of Commerce and established the Artificial Intelligence Learning Laboratory Program, which the office administers. Through the program, OAIP and the relevant state agency may grant a participant temporary relief from specified state rules under a “regulatory mitigation agreement.” In each healthcare pilot, the Division of Professional Licensing agrees not to bring enforcement actions under specified provisions, chiefly the prohibitions on unlicensed practice and unprofessional conduct and certain telehealth requirements, while the participant follows the safeguards set out in its agreement.

The relief is limited. Under the statute, as amended in 2026, an initial demonstration period may not exceed 12 months, and the office may grant up to two 12-month extensions. Participants send OAIP monthly reports and must report adverse events, and the office can demand a remediation plan or end an agreement at any time. Participants remain subject to every law their agreement does not expressly waive, and the agreements waive only specified Utah provisions. The statute also provides that the state is not responsible for claims arising from a participant’s involvement in the program, and that signing an agreement “does not constitute an endorsement or approval from the state.”

The Pilots

Utah’s healthcare agreements fall into three groups: AI renewal of existing prescriptions, AI issuance of new prescriptions, and agreements that set the stage for future pilots. Status information below reflects the state’s authorized pilots page as of November 9, 2026.

Group 1: Renewal of Existing Prescriptions

Doctronic (maintenance medications). Utah’s first AI prescription renewal agreement was signed in October 2025 with Doctronic, a digital health company that markets an AI “doctor.” The state publicly announced the pilot on January 6, 2026. Under the agreement, the AI may process only 30-, 60-, or 90-day renewals of medications that a licensed provider has already prescribed, drawn from a fixed list of about 190 maintenance medications in categories such as blood pressure, diabetes, and mental health. Doctronic describes the service as aimed at patients whose prescriptions have run out of refills. The AI may not issue new prescriptions, change a treatment plan, or renew controlled substances.

The agreement sets three phases of oversight: licensed physicians review each AI renewal decision before it is sent to the pharmacy, then review decisions after the fact, and finally review a sample of renewals. OAIP later revised the path forward so that each medication group must reach 250 filled prescriptions before it can advance, and moving to the second phase still requires OAIP approval. As of November 9, 2026, the state’s authorized pilots page listed the pilot as active and still in the first phase, in which a licensed medical practitioner authorizes every request. The agreement’s term ran from October 24, 2025, to October 24, 2026, and allows a single extension of up to 12 months. The state’s website does not indicate whether it had been extended.

Doctronic is the only participant whose results the state has published. According to OAIP’s May 2026 report, which covers January through April 2026, the AI recommended renewal in 72% of cases and sent the other 28% to a physician because of the characteristics of those cases. Reviewing physicians agreed with 91% of the AI’s renewal recommendations (97% when a second physician’s opinion was considered) and agreed that 69% of the escalations were warranted. The report notes that these figures are based on reports from Doctronic’s physicians. It also states that OAIP had initiated a separate, independent review of the interactions.

Legion Health (psychiatric medication). Utah’s second renewal agreement, with Legion Health, was signed on March 19, 2026. According to the state’s pilot page, Legion’s AI may renew existing, non-controlled psychiatric medications, such as common treatments for anxiety and depression, for stable patients. It may not diagnose conditions, start new medications, or change doses. A licensed clinician is to review the first 250 requests before they are completed, the next 1,000 are subject to retrospective review, and the pilot then moves to monthly random sampling. The state’s authorized pilots page lists the agreement as approved and states that its demonstration period had not started.

August AI (maintenance medications). August’s agreement, announced October 5, covers refills of existing, non-controlled maintenance medications for patients with an established treatment plan, including drugs for diabetes, hypertension, and high cholesterol. The state’s authorized pilots page listed the agreement, signed October 2, as approved, with its demonstration period not yet started. Medications are grouped into cohorts, and each cohort moves through three phases: a Utah-licensed healthcare provider reviews the first 250 AI refill decisions before they reach patients, the next 1,000 are reviewed after the fact, and then at least 5% of prescriptions in each medication class are spot-checked. Each new phase requires OAIP’s prior written approval.

Group 2: Initial Prescriptions

Nolla Health (topical acne treatment). Nolla’s pilot, with a term of October 5, 2026, to October 5, 2027, is the only prescribing pilot on the state’s list that authorizes first-time prescriptions. Adults with mild to moderate acne complete a medical intake and facial photo scans, and the AI chooses from a small, predefined formulary of topical treatments. Oral medications, isotretinoin, and severe acne are excluded. Under the agreement, two Utah-licensed physicians review every prescription before it is sent during the first stage, which covers at least the first 100 patients. In later stages, the AI submits prescriptions directly while a physician reviews every case after the fact and, eventually, a monthly sample of at least 10% of cases, with any case involving an adverse event or escalation still reviewed in full. The first stage must last at least four weeks and the second at least eight weeks, and each step requires minimum cumulative enrollment (500 and 750 patients for the later stages), at least 95% alignment with physician decisions, no missed safety “hard stops,” no serious adverse events, and OAIP’s written approval. Nolla’s press release says the company “met and worked directly with the Utah Medical Board” before launch.

Group 3: Other Agreements

The state signed a pilot agreement with Expect Fitness on October 2, 2026, which permits the company’s AI to help create physical therapy plans for urinary incontinence, pelvic pain, and postpartum pelvic floor conditions. In the first phase, a licensed physical therapist reviews every plan and may change or reject it.

Utah has also entered into master agreements with the University of Utah Health and Intermountain Health which set standing terms for future AI pilots that the health systems can propose. Each pilot, which must be approved in a separate written addendum to the master agreements, would run for 12 months, with up to two 12-month extensions. As of November 9, no pilots had been approved under either agreement.

Questions About Evidence and Oversight

The program has drawn criticism from within Utah’s medical community. On April 20, 2026, the Utah Medical Licensing Board sent a letter signed by 11 of its 14 Board members, recommending that the Doctronic pilot be suspended immediately. The Board wrote that it learned of the agreement only after the system was already live, and that each refill requires “reassessment and clinical decision-making.” “There is a reason prescription refills require physician authorization,” the Board wrote.

The next day, the Division of Professional Licensing and OAIP declined to suspend the pilot. They noted that a licensed physician reviews and approves every AI-generated renewal during the first phase and that OAIP retains authority to modify or cancel the pilot if safety benchmarks are not met. The agencies invited the Board to review pilot data with them, and on June 18, 2026, OAIP Director Zach Boyd presented to the Board and asked for its help in assessing how current pilots are performing and evaluating new pilot proposals.

In its October 5 announcement, the Department of Commerce also said the state had partnered with six outside organizations, which OAIP’s evaluator page places in three groups. Stanford’s Clinical Excellence Research Center, mpathic, and Vega Health conduct clinical and technical evaluations, which can include auditing samples of an AI system’s activity against the company’s own reports. Clarion AI Partners and Glacis Technologies provide runtime attestation: signed receipts showing that a specified safety check ran, which, as the page notes, do not show “that the AI is safe or that the company followed the law.” The Coalition for Health AI qualifies partner evaluators rather than evaluating pilots itself. Companies pay their evaluators directly for evaluation engagements; the state does not pay for evaluations, and attestation receipts are free to participating companies.

The newer agreements also require participants to tell users when they are interacting with generative AI, maintain professional liability insurance, and report adverse events promptly (within 24 hours under the Nolla and August agreements, and by the next business day for serious events under the Expect Fitness agreement). Each agreement also provides that it does not displace users’ existing legal remedies.

Federal Oversight

Utah’s sandbox agreements waive only specified state requirements, and participants remain subject to federal law. FDA regulates software that meets the definition of a medical device, and the 21st Century Cures Act excludes certain clinical decision support (CDS) software from that definition. Under the 21st Century Cures Act, as interpreted in FDA’s CDS guidance (issued January 2026), software qualifies for the exclusion only if it meets all four criteria: it does not acquire, process, or analyze medical images or signals; it displays, analyzes, or prints medical information, such as patient records or clinical practice guidelines; it supports or provides recommendations to a healthcare professional; and it enables that professional to independently review the basis for those recommendations rather than rely primarily on them. Physician review of AI output is relevant to that analysis but does not by itself establish the exclusion. Under the guidance, software intended for patients or caregivers rather than clinicians falls outside the exclusion, as does software that analyzes images for a medical purpose, including images not originally taken for one. One may reasonably ask whether Utah’s pilots fit within the CDS exclusion. Nolla Health’s tool, for example, analyzes facial photographs submitted by patients, along with other patient information, to select an acne treatment. That image analysis appears to be in tension with the first criterion discussed above. In the later stages, if the state approves them, the tool would send prescriptions to the pharmacy before a physician reviews them, which appears to be in tension with the fourth criterion.

Legislators are making attempts to clarify these issues. In Congress, the Healthy Technology Act of 2025 (H.R. 238) would let AI qualify as a prescriber if authorized by state law and approved, cleared, or authorized by FDA. The bill has not moved beyond referral to the House Energy and Commerce Committee. The regulatory ambiguity has not deterred Doctronic, which said in March 2026 that it was in discussions with regulators in other states and governments about expanding beyond Utah.

The Next Question: When Should Oversight Step Down?

Each of Utah’s prescribing pilots is designed to reduce physician involvement over time, moving from approval of every prescription in advance to review after the fact and, eventually, to sampling. Whether each step is justified will depend on the evidence the pilots produce, such as concordance with physician decisions, escalation rates, and adverse event reports, and on whether the independent evaluators find that evidence reliable.

The Utah tools are narrowly scoped; the respective agreements limit the permitted uses to defined medications or treatment protocols. But as AI agents take on more work in healthcare and elsewhere, there may be pressure to let tools like these do more, such as ordering labs, scheduling follow-up, or coordinating with pharmacies, with less human involvement.

As our colleagues Michael Adler and Scott Meyers explain in a recent Bloomberg Law article, “configuration matters alongside the model: what agents can reach, what they’re told, and what can stop them.” Human review is a meaningful control only when reviewers have the information and authority to act on what they see. As the Utah pilots move toward less frequent review, and as tools like these may be asked to do more, the evidence supporting each step and the safeguards that remain might determine whether these pilots become a model for other states or a cautionary tale. Akerman will continue to follow and report on these efforts.

  • Posted in:
    Health Care
  • Blog:
    Health Law Rx
  • Organization:
    Akerman LLP
  • Article: View Original Source

LexBlog logo
Copyright © 2026, LexBlog. All Rights Reserved.
Legal content Portal by LexBlog LexBlog Logo