When AI Writes the Prescription: Licensure, Liability, and the Corporate Practice of Medicine in Utah’s Autonomous AI Pilot

By Tony Caldwell and Scott Driggs

In the first known instance in the United States, an artificial intelligence (AI) system is autonomously prescribing medication to patients without a physician in the loop. Utah’s Medical Licensing Board has demanded immediate suspension of the pilot activities even as the program continues beyond the pilot phase and its developer, Doctronic, eyes expansion into other states. The dispute is a developing story that may shape the future of autonomous AI in healthcare delivery.

This alert summarizes the Utah autonomous AI pilot’s legal architecture and principal pressure points, including licensure, liability, Food and Drug Administration (FDA) device status, and corporate practice of medicine. Healthcare organizations, practitioners, regulators, and technology developers in Utah and beyond are closely studying the pilot as a bellwether for the transition from clinical decision support to autonomous AI action.

On January 6, 2026, the Utah Office of Artificial Intelligence Policy (OAIP) publicly launched Doctronic’s autonomous AI prescription pilot under Utah’s Artificial Intelligence Learning Laboratory Program (the AI Prescription Pilot).¹ The program, created by Utah’s Artificial Intelligence Policy Act, Utah Code Title 13, Chapter 72,² functions as a regulatory “sandbox.” It permits companies to test innovative AI products in the real world for a limited time, under state supervision, with temporary relief from specified state rules. Each participant operates under a regulatory mitigation agreement that the OAIP negotiates with the relevant state agency. The agreement can waive or modify specified state enforcement, but it is not a general license to practice medicine and cannot waive federal law.³ The AI Prescription Pilot permits automated, guideline-based renewals of medications previously prescribed by a licensed provider, subject to the AI Prescription Pilot’s stated safeguards and Utah regulatory oversight. Doctronic is also in active discussions with regulators in Arizona, Texas, and Wyoming about similar programs.⁴

Scope of the AI Pilot. The AI Prescription Pilot covers prescription renewals for 191 medications for chronic conditions, including hypertension, diabetes, and depression. The AI Prescription Pilot may process 30-, 60-, or 90-day renewals of existing prescriptions, but may not issue new prescriptions, handle controlled or addictive substances, or change treatment plans.

Phased implementation. The AI Prescription Pilot remains in Phase One. The OAIP has since tightened the Phase Two criteria: the original agreement allowed the AI Prescription Pilot to advance after 250 total prescriptions were filled, but each medication group must now independently reach 250 filled prescriptions, and any transition remains subject to the OAIP approval.

Early results. The OAIP’s May 2026 report covering January through April found that the AI recommended renewal in 72% of cases and escalated the remaining 28% to a physician. 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 escalations were warranted, characterizing the remainder as overly cautious. The data was reported by Doctronic’s own physicians, and the OAIP has initiated a separate, independent review.⁵

The underlying regulatory mitigation agreement entered into by and between Doctronic, the OAIP, and the Utah Department of Professional Licensing was executed for a 12-month term running from approximately October 2025 through October 2026 (although the system did not go live until the January 2026 public launch). By statute, an initial agreement may run no more than 12 months; a participant may request an extension no later than 30 days before the term ends, and no more than two 12-month extensions may be granted.⁶ Participants must also promptly report consumer harm, privacy breaches, and unauthorized data use, which may lead to removal from the program.⁷ A second AI company, Legion Health, has signed an agreement with the OAIP for a similar renewal pilot limited to non-controlled psychiatric medications, although its demonstration period has not yet begun.⁸

The AI Prescription Pilot tests how Utah’s Medical Practice Act applies when an AI system makes the initial clinical determination to renew a prescription rather than a licensed physician.

Licensure requirement. Utah Code § 58-67-301 requires a license to practice medicine in Utah.⁹ Deciding whether a prescription remains appropriate can require clinical judgment about condition, contraindications, interactions, and treatment effectiveness. The sandbox approval of the AI Prescription Pilot does not resolve whether an AI system performing that function is engaging in the unlicensed practice of medicine. Utah’s separate generative AI disclosure law likewise treats certain personalized medical advice or services as high-risk interactions.¹⁰

Corporate practice of medicine. Utah Code § 58-67-802 permits practice of corporate medicine through specified business forms but requires the practice to be conducted by an individual who is licensed and competent to practice medicine.¹¹ That language creates a direct question for an AI system, such as Doctronic’s AI tool utilized in the AI Prescription Pilot, that cannot hold a medical license.

The statutory reference to an “individual” supports the argument that human licensure remains central, even where the state of Utah has authorized a limited pilot without human licensure.

Unlawful conduct. Utah’s unlawful-conduct provisions address interference with a medical licensee’s lawful practice and direct clinical care by nonphysicians.¹² Today, that concern is limited because a physician reviews and authorizes every renewal. If the AI Prescription Pilot advances to Phase Two, however, the AI would send renewals directly to pharmacies, and physicians would no longer authorize each renewal in advance. At that point, it becomes easier to argue that the AI system, rather than a licensed physician, is directing patient care.

Medical Board opposition. In April 2026, 11 of 14 physicians on the Utah Medical Licensing Board signed a letter calling for the AI Prescription Pilot’s “immediate suspension.”¹³ The Board criticized the state for launching the program without consulting it and emphasized that each prescription renewal requires reassessment and clinical decision-making. The OAIP and the Division of Professional Licensing responded by forming a joint working group with the Board.¹⁴ Public Citizen, a national nonprofit, sent a May 2026 letter supporting suspension.¹⁵ A New England Journal of Medicine Perspective by Sara Gerke, Ravi Parikh, and I. Glenn Cohen and a Health Affairs analysis by Weissman characterized Utah’s approach as a “flawed regulatory playbook.”¹⁶

The liability framework for autonomous clinical AI remains largely untested. Traditional malpractice, product liability, contract, and professional-discipline theories may all apply, but Phase One of the AI Prescription Pilot and any future Phase Two would present materially different risk profiles.

Sandbox participation does not shift that risk to the state. Utah disclaims responsibility for claims, liabilities, damages, losses, or expenses arising from a participant’s involvement in the program.¹⁷

Learned intermediary problem. In Phase One, a physician reviews each recommendation and may serve as the independent medical intermediary contemplated by traditional product-liability doctrine. That analysis is less certain when the AI acts without physician review.

Phase Two would remove the prescribing physician from the immediate decision chain. A pharmacist may provide a check, but the shift could increase arguments that the developer or deployer owes duties directly to the patient.

Contractual risk allocation. Reported arrangements include malpractice coverage for Doctronic physicians and broad vendor terms disclaiming responsibility for system accuracy or harmful patient outcomes.

Product liability. The outcome in Garcia v. Character Technologies, Inc. suggests that courts may consider AI systems products for design-defect claims.¹⁸ An autonomous prescribing system could therefore face product-liability theories independent of malpractice. Courts may scrutinize a provider or vendor that holds itself out as delivering medical services while disclaiming responsibility for the accuracy of those services, particularly where patients have limited bargaining power.

State attorneys general have also signaled increasing scrutiny of AI systems associated with patient harm, suggesting that reduced physician oversight may attract both private claims and public enforcement.¹⁹

Physicians whose names appear on AI-generated prescriptions may bear personal exposure for decisions they did not individually make. Insurance terms and contracts should allocate responsibility expressly among physicians, developers, and platform operators.

Sandbox relief does not resolve the federal question. Participants remain subject to legal and regulatory requirements not expressly waived or modified by the agreement, including applicable federal law.²⁰ A central issue is therefore whether the Doctronic system is a medical device subject to FDA oversight or exempt clinical decision support software under the 21st Century Cures Act.

Clinical decision support. Section 520(o) of the 21st Century Cures Act generally excludes software that supports or recommends care to healthcare professionals and enables them to independently review the basis for the recommendation.²¹ Phase One may fit that model because a physician reviews every recommendation from the AI Prescription Pilot.

Autonomous action and evolving FDA engagement. Phase Two would be harder to reconcile with the clinical decision support exemption because the AI system would submit renewals directly to pharmacies. In August 2026, FDA issued its most direct engagement with generative AI-enabled medical devices to date, a discussion paper addressing risk assessment, premarket evaluation, and postmarket monitoring.²² The paper is not final guidance, but it underscores that autonomous systems remain an unsettled regulatory category. H.R. 238, the pending Healthy Technology Act of 2025, would amend the Federal Food, Drug, and Cosmetic Act (FD&C) Act to allow AI systems to prescribe medications if authorized by state law and approved, cleared, or authorized by FDA, further suggesting that the current legal picture is unsettled.²³

If an AI system is treated as a medical device, the consequences may include FDA premarket review, quality system requirements, adverse-event reporting, and enforcement exposure for operation without authorization.

The AI Prescription Pilot serves as a good case study for the future of autonomous AI in healthcare. Regulators, medical boards, courts, legislatures, and healthcare organizations are watching it as a bellwether for how additional autonomous clinical tools may be governed. The issue is evolving and is being actively studied across the healthcare community. Doctronic’s $40 million Series B in March 2026²⁴ and its acquisition of pediatric care company Summer Health signal that investors see a viable market and desire expansion well beyond routine prescription renewals.

However, published commentary, including the Gerke, Parikh, and Cohen Perspective and the Health Affairs analysis describing Utah’s approach as a “flawed regulatory playbook,” have highlighted limited pre-deployment evidence, unresolved accountability, and the risk that automated renewals may eliminate preventive-care touchpoints.

A contrasting approach. Other states are moving to keep humans in the loop rather than testing their removal. Colorado’s SB 26-189, effective January 1, 2027, gives consumers the right to request meaningful human review after an adverse decision made using automated decision-making technology.²⁵ Colorado’s HB 26-1139 requires that AI-driven coverage denials based on medical necessity be reviewed by a licensed clinician, and it bars payers from paying for psychotherapy conducted by AI.²⁶ Alabama, Iowa, and Washington have likewise restricted AI’s role in utilization review.²⁷ For developers pursuing multistate growth, the lesson is that a Utah sandbox agreement does not travel: the same autonomous function that Utah permits on a trial basis may face mandatory human-review requirements elsewhere. For a fuller discussion of Colorado’s framework, see our alert, Update: Colorado’s Three-Front AI Healthcare Landscape Has Shifted.

Before deploying AI that acts without physician sign-off, health systems, provider groups, and developers should consider the following: