
Photo: Shutterstock
A Utah pilot program would allow AI to write prescriptions, but the American Medical Association says AI shouldn’t practice medicine

Photo: Shutterstock
For several years, AI tools have been rolled out to help support doctors. Now, Utah has launched a pilot program that would, for the first time in the U.S., allow AI to issue new prescriptions — in this case, for acne.
The program is in its early stages, with doctors still reviewing and approving every prescription. But in later stages, the AI program would act autonomously, reviewing images of a person’s face and writing a prescription for acne medication.
The development comes as physicians are drawing new boundaries around AI’s role in medical care, with a new policy from the American Medical Association (AMA) arguing that writing prescriptions is an essential part of the practice of medicine.
AI can support physicians, AMA writes, “but don’t call it doctor.”
In Utah, a company called Nolla Health is running the pilot under a 12-month agreement with the state’s Office of Artificial Intelligence Policy. Adults in the state with mild to moderate acne can complete a questionnaire and face scan, and the system selects from pre-approved topical treatment plans, Nolla says.
In the first phase of the program, two physicians review each prescription before it reaches the patient. In the second, the AI submits prescriptions directly, and a physician reviews them afterward, at least once a week. In the third, physicians review a sample of at least 10% of prescriptions each month, plus any case involving an adverse event or escalation. Advancement to the next phase requires at least 95% agreement between the AI and physicians, no serious adverse events, and the state’s written approval.
Nolla says the program could help address gaps in access to dermatological care. The company says roughly 100,000 Utahns have active acne, only about 10% see a dermatologist, the average wait is 61 days, and 11 counties have no dermatologist.
John Whyte, M.D., CEO of the AMA and a board-certified internist, says the pilot’s endpoint is the problem: “We believe that prescribing medications is an inherent part of the practice of medicine, and that these tools are not licensed to practice medicine.”
Eventually, the goal of the program is “to prescribe medications autonomously,” he adds. “We wholeheartedly reject that from a safety perspective.”
The concern reaches beyond U.S. physicians. Henrique Dias, cofounder of NoHarm, a Brazilian nonprofit that uses AI to check hospital prescriptions, told The BMJ that deploying autonomous AI to prescribe without clinician oversight is “the wrong trajectory for healthcare.”
Beyond concerns about autonomous prescribing, physicians also question whether the technology can account for differences in how skin conditions present.
Ade Adamson, M.D., a dermatologist at the University of Texas at Austin, says acne treatment may also need to account for hyperpigmentation. “I have yet to see data that supports this AI agent being able to sift through those differences.”
“We [also] know that in people of color, acne often presents differently. It could be rosacea, it could be another condition,” Dr. Whyte adds.
While the pilot program will only prescribe topical medication — meaning the stakes are lower than, for example, a blood pressure pill — they aren’t zero, Dr. Adamson notes. “You may have scarring. You may have a delay in the actual treatment that you need.”
Follow-up also matters, adds Dr. Whyte: “There’s no longitudinal care as part of these tools, either. So who’s following up?”
Dr. Adamson’s biggest objection is that no randomized trial is under way. “Why are they not running a randomized controlled trial of this intervention? That is the gold standard for figuring out whether this works or not,” he says. Digital health companies, he adds, often “just roll out a program and just assume it’s going to be better because it’s AI.”
The state agreement uses physician consensus with the AI as the pilot’s primary safety metric, rather than patient outcomes. Nolla says clinicians agree with its recommendations in 96% of real-world cases, but its website doesn’t explain how it arrived at that figure.
Utah’s Office of AI Policy says pilots remain under strict oversight and that independent evaluators will verify developers’ claims. But the AMA’s new policy calls for “transparent, auditable evidence demonstrating their safety and efficacy” before AI tools are used for prescribing.
“Since when in health do we say ‘let’s go try something,’” asks Dr. Whyte, “as opposed to having some data that it works?”
DISCLAIMER: MedShadow provides information and resources related to medications, their effects, and potential side effects. However, it is important to note that we are not a substitute for professional medical advice, diagnosis, or treatment. The content on our site is intended for educational and informational purposes only. Individuals dealing with medical conditions or symptoms should seek guidance from a licensed healthcare professional, such as a physician or pharmacist, who can provide personalized medical advice tailored to their specific circumstances.
While we strive to ensure the accuracy and reliability of the information presented on MedShadow, we cannot guarantee its completeness or suitability for any particular individual’s medical needs. Therefore, we strongly encourage users to consult with qualified healthcare professionals regarding any health-related concerns or decisions. By accessing and using MedShadow, you acknowledge and agree that the information provided on the site is not a substitute for professional medical advice and that you should always consult with a qualified healthcare provider for any medical concerns.
Help fund our fight for improved medication safety and transparency by supporting our independent journalism.
Give Now
Help fund our fight for improved medication safety and transparency by supporting our independent journalism.
Give NowWe seek to create a world where there are safer medications for all and where all people are fully informed about their medication benefits and side effects, using our explanatory and investigative journalism to both inform and stimulate public discussion to drive toward solutions.
MedShadow.org is dedicated to empowering people with our evidence-based journalism about the safest ways to take over-the-counter (OTC) and prescription (Rx) medications.
We also seek to stimulate public discussion about potential improvements to drug research, manufacturing and distribution processes. and also to encourage the improvement of processes to ensure that we are all better protected.
As a result, we encourage other nonprofit newsrooms, mission-aligned outlets, and advocacy organizations to republish our original content under the following conditions:
1. We Are Credited Fully and Clearly
You must include the byline and credit MedShadow.org as the original publisher. Please include a link to the original article at the top or bottom of the piece. Use the following attribution:
This article was originally published by MedShadow, a nonprofit foundation that focuses on the safest ways to take over-the-counter and prescription medications.
2. Our Content Isn’t Edited Without Permission
You may make small changes for clarity, such as adjusting headlines, subheadings, or transitions, but you may not change the core meaning, tone, or intent of the article. For substantive edits, email us at republish@medshadow.org for review and approval.
3. You Do Not Sell Or Use Our Content Commercially
Our work is free to share, but it may not be sold, monetized, or repurposed for commercial gain. You may not republish our articles behind a paywall or in publications whose primary purpose is advertising or product promotion. You may not place drug or supplement advertising adjacent to our story.
4. Photos, Graphics, and Multimedia Are Not Used Without Permission
Only the text of the article is available for republication. Images, graphics, and videos may not be reused without explicit written permission, unless they are clearly marked as Creative Commons licensed. Contact us if you’d like to request reuse of a visual asset.
If you’re unsure whether your intended use falls within these guidelines, or would like to discuss a formal syndication or licensing arrangement, please email: republish@medshadow.org
We want our journalism to reach as many people as possible, especially those making critical decisions about medications and health.
For website editors
Copy the article body, including its formatting and links, then paste it into your publishing system. Add the byline and required credit described in these guidelines.
<!-- wp:paragraph --> <p>For several years, AI tools have been rolled out to help support doctors. Now, Utah has launched a pilot program that would, for the first time in the U.S., allow AI to issue new prescriptions — in this case, for acne. </p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>The program is in its early stages, with doctors still reviewing and approving every prescription. But in later stages, the AI program would act autonomously, reviewing images of a person’s face and writing a prescription for acne medication. </p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>The development comes as physicians are drawing new boundaries around AI's role in medical care, with a new policy from the American Medical Association (AMA) arguing that writing prescriptions is an essential part of the practice of medicine. </p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>AI can support physicians, AMA writes, “<a href="https://www.ama-assn.org/press-center/ama-press-releases/ama-artificial-intelligence-can-augment-physicians-don-t-call-it">but don’t call it doctor.</a>” </p> <!-- /wp:paragraph --> <!-- wp:heading --> <h2 class="wp-block-heading"><strong>How Utah's AI Prescription Pilot Works</strong></h2> <!-- /wp:heading --> <!-- wp:paragraph --> <p>In Utah, a company called Nolla Health is running the pilot under a <a href="https://commerce.utah.gov/wp-content/uploads/2026/10/RMA-Nolla-Health.pdf">12-month agreement</a> with the state’s Office of Artificial Intelligence Policy. Adults in the state with mild to moderate acne can complete a questionnaire and face scan, and the system selects from pre-approved topical treatment plans, <a href="https://www.nollahealth.com/blog/ai-prescriptions-utah">Nolla says</a>.</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>In the first phase of the program, two physicians review each prescription before it reaches the patient. In the second, the AI submits prescriptions directly, and a physician reviews them afterward, at least once a week. In the third, physicians review a sample of at least 10% of prescriptions each month, plus any case involving an adverse event or escalation. Advancement to the next phase requires at least 95% agreement between the AI and physicians, no serious adverse events, and the state's written approval.</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>Nolla says the program could help address gaps in access to dermatological care. The company says roughly 100,000 Utahns have active acne, only about 10% see a dermatologist, the average wait is 61 days, and 11 counties have no dermatologist. </p> <!-- /wp:paragraph --> <!-- wp:heading --> <h2 class="wp-block-heading"><strong>Why Doctors Are Concerned About AI Prescriptions</strong></h2> <!-- /wp:heading --> <!-- wp:paragraph --> <p><a href="https://www.ama-assn.org/about/authors-news-leadership-viewpoints/john-j-whyte-md-mph">John Whyte, M.D.</a>, CEO of the AMA and a board-certified internist, says the pilot's endpoint is the problem: "We believe that prescribing medications is an inherent part of the practice of medicine, and that these tools are not licensed to practice medicine." </p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>Eventually, the goal of the program is “to prescribe medications autonomously,” he adds. “We wholeheartedly reject that from a safety perspective."</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>The concern reaches beyond U.S. physicians. <a href="https://www.linkedin.com/in/hdpsantos?originalSubdomain=br">Henrique Dias</a>, cofounder of NoHarm, a Brazilian nonprofit that uses AI to check hospital prescriptions, told <a href="https://www.bmj.com/lookup/doi/10.1136/bmj-2026-101186">The BMJ</a> that deploying autonomous AI to prescribe without clinician oversight is "the wrong trajectory for healthcare."</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>Beyond concerns about autonomous prescribing, physicians also question whether the technology can account for differences in how skin conditions present.</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p><a href="https://healthcare.ascension.org/find-care/provider/1255630034/adewole-adamson">Ade Adamson, M.D</a>., a dermatologist at the University of Texas at Austin, says <a href="https://medshadow.org/conditions-treatments/skin-conditions/is-benzoyl-peroxide-still-safe-what-to-know-about-acne-treatment-recalls/">acne treatment</a> may also need to account for hyperpigmentation. "I have yet to see data that supports this AI agent being able to sift through those differences."</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>"We [also] know that in people of color, acne often presents differently. It could be rosacea, it could be another condition," Dr. Whyte adds.</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>While the pilot program will only prescribe topical medication — meaning the stakes are lower than, for example, a blood pressure pill — they aren’t zero, Dr. Adamson notes. "You may have scarring. You may have a delay in the actual treatment that you need." </p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>Follow-up also matters, adds Dr. Whyte: "There's no longitudinal care as part of these tools, either. So who's following up?"</p> <!-- /wp:paragraph --> <!-- wp:heading --> <h2 class="wp-block-heading"><strong>Is There Enough Evidence That AI Prescribing Is Safe?</strong></h2> <!-- /wp:heading --> <!-- wp:paragraph --> <p>Dr. Adamson's biggest objection is that no randomized trial is under way. "Why are they not running a randomized controlled trial of this intervention? That is the gold standard for figuring out whether this works or not," he says. Digital health companies, he adds, often "just roll out a program and just assume it's going to be better because it's AI."</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>The state agreement uses physician consensus with the AI as the pilot's primary safety metric, rather than patient outcomes. Nolla says clinicians agree with its recommendations in 96% of real-world cases, but its website doesn't explain how it arrived at that figure.</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>Utah's Office of AI Policy says pilots remain under strict oversight and that independent evaluators will <a href="https://commerce.utah.gov/2026/10/05/utah-enhances-pro-human-ai-initiative-with-new-healthcare-pillar-adds-new-pilots-to-its-regulatory-sandbox-and-partners-with-third-party-evaluators/">verify</a> developers' claims. But the AMA's new policy calls for "transparent, auditable evidence demonstrating their safety and efficacy" before AI tools are used for prescribing.</p> <!-- /wp:paragraph --> <!-- wp:paragraph --> <p>"Since when in health do we say ‘let's go try something,’” asks Dr. Whyte, “as opposed to having some data that it works?"</p> <!-- /wp:paragraph -->
We think it’s invaluable — and hope you agree. Our work is made possible by donations from readers like you.
We think it’s invaluable — and hope you agree. Our journalism is made possible by donations from readers like you.
As a nonprofit, we don’t take a dime from pharmaceutical or supplement companies. That means we can report on the benefits, risks, and alternatives to prescription and over-the-counter drugs free from corporate pressure or influence.
We've noticed that you're interested in this topic. Would you read a dedicated newsletter about it?