Utah's AI Prescriber Has One Safety Study. Every Author Profits If It Succeeds.
Allowing an AI model to renew prescriptions seems like a straightforward way to make healthcare more accessible. Obtaining a prescription renewal in the U.S. can take weeks and some patients even turn to the emergency room to access medications. An AI prescription service could solve these problems. However, the legal framework currently used in Utah is dangerously ambiguous. It could harm patients and create pushback against AI in healthcare.
Doctronic, the AI company approved to issue prescription renewals in Utah, is operating in a legal gray zone. Under federal law only a licensed medical practitioner can issue a prescription. The word ‘practitioner’ has always implied a human being. While Utah waived state licensure laws for Doctronic under its AI regulatory sandbox, the federal legal status of the program remains unresolved.1 The FDA has not authorized the system, and has not publicly commented on the pilot. The legal ambiguity would be less concerning if the AI itself were provably safe. But early testing remains inconclusive. The 500-patient study, which claims to show 99.2% consistency between AI and physician decisions, was conducted by Doctronic’s own research group. Every author of the study is a financial stakeholder in Doctronic.2 This alone should raise concerns, but the study itself is poorly designed.
The most notable design flaw is the benchmark set for the AI system’s performance. Doctronic’s decisions were compared to clinician decisions rather than correct decisions. This is problematic because the practitioner and AI system could both make an incorrect judgement and it would appear as though the AI performed well. This is especially concerning because there is no way to know whether the medical professionals participating in the study are representative of physicians.
The Doctronic-sponsored study does not specify the number of clinicians participating, their specialties, or their years of practice. The study describes its comparator group only as “board-certified, U.S.-licensed clinicians that specialize in telehealth”. This vague language suggests that nurse practitioners or physician assistants may have been included in the research. As a result the assumption that clinicians participating in the study are representative of physicians is dubious. A study with both conflict of interest and an uncharacterized comparator group should not be used to inform public policy decisions.
This bias and poor design extends beyond early research. The pilot program itself uses physicians employed by Doctronic to oversee the AI, another clear conflict of interest. The oversight itself is also problematic because of its subjective nature. Any cases of disagreement between the automated system and the reviewing clinician are “further reviewed, analyzed, and assessed for risk using an established academic scoring procedure,” with reported cases categorized as “no risk,” “minor risk,” or “minor-to-moderate risk.”3 There’s no public definition of what constitutes each category, who applies that scoring, or what the escalation pathway looks like when something is flagged as moderate risk. In fact, that scoring procedure is defined inside Doctronic’s own regulatory proposal, a document the state of Utah has classified as a protected record and shielded from public records requests. Utah citizens cannot independently verify what safety standards their AI prescriber is actually being held to.
When the AI does make a mistake, the liability structure offers little reassurance. It is not clear if Doctronic or the clinician whose name is on the prescription is at fault. Traditional malpractice law suggests that the physician is at fault. However, under Utah’s current pilot program that provider may never meet with the patient or review the case. Instead, in phase III of the pilot program only 5-10% of AI issued prescriptions will be reviewed by a physician at all. The signed agreement explicitly permits a named prescriber to fulfill their oversight obligations without seeing the patient.
AI integration into healthcare should not be rushed. Lack of transparency and poorly designed research have contributed to serious public health disasters in the past. These include an opioid epidemic driven by company-sponsored research that deliberately minimized addiction risk, and an HIV-contaminated blood supply that infected more than half of the 16,000 hemophiliacs in the United States.45 AI should not repeat these mistakes. It has the power to make healthcare more accessible and efficient. Instead, the current path appears to be low standards for research and disregard for patient safety.
This rhetoric may seem extreme for a pilot program with limited scope. It is not. Standards set during early pilots become the template for what follows. If states accept company-sponsored research and ambiguous liability policies as sufficient grounds for deploying autonomous AI prescribers, those same standards will be cited when the next company applies. This next company could include higher-risk medications and larger populations. The question is not just whether Doctronic will harm someone in Utah this year. It is what precedent we are setting for the next decade of AI in medicine. Rather than allowing individual states to race ahead under regulatory sandboxes, the federal government should establish minimum evidentiary standards. At a bare minimum, research with publicly available methods and without flagrant conflict of interest should be required before any AI system is deployed in healthcare. AI could be a powerful tool to improve patient outcomes and accessibility, but it could also cause harm and create distrust. It is important that governments protect patients by using sound research to form policy even if it means slower deployment of AI systems.
21 U.S.C. § 353(b)(1) — Federal Food, Drug, and Cosmetic Act, Prescription Drug Provisions. The statute states that prescription drugs “shall be dispensed only upon a written prescription of a practitioner licensed by law to administer such drug,” and that dispensing a drug contrary to this provision “shall be deemed to be an act which results in the drug being misbranded.”
Doctronic Research Group, "Toward the Autonomous AI Doctor," medRxiv preprint, July 2025. Full text. Note: All authors are equity owners of Doctronic Inc. Ethics approval was granted by Doctronic's own internal ethics committee.
Utah Department of Commerce, Office of Artificial Intelligence Policy. AI Doctronic — Authorized AI Pilots. Accessed June 2026. https://commerce.utah.gov/ai/regulatory-relief/authorized-ai-pilots/doctronic/
U.S. Department of Justice. Opioid Manufacturer Purdue Pharma Pleads Guilty to Fraud and Kickback Conspiracies. November 24, 2020. https://www.justice.gov/archives/opa/pr/opioid-manufacturer-purdue-pharma-pleads-guilty-fraud-and-kickback-conspiracies
Institute of Medicine. HIV and the Blood Supply: An Analysis of Crisis Decisionmaking. National Academies Press, 1995.
National Academies Press (full text): https://nap.nationalacademies.org/read/4989/chapter/2
NIH/NCBI hosted version: https://www.ncbi.nlm.nih.gov/books/NBK232413/