Utah broke ground in January. It became the first state to allow an autonomous AI system to renew prescriptions. The move didn’t just test technology; it tested the boundaries of medical licensing, regulatory power, and public trust. Now, the American Medical Association, the state medical board, and the FDA are all shouting at each other over what constitutes safe care.
The core agreement involves Doctronic and the state’s Office of Artificial Intelligence Policy. Under this deal, Utah agreed not to enforce its unprofessional-conduct laws against the company. In exchange, Doctronic adheres to strict safety and privacy guardrails. This regulatory workaround is the heart of the “Utah AI prescribing pilot” — a program that lets a machine handle routine medical maintenance without a doctor’s immediate oversight.
How the Utah AI Prescription System Actually Works
If you have a chronic condition, the process is straightforward. Doctronic’s AI can refill 30-, 60-, or 90-day supplies. The cost is $4 per prescription. It covers roughly 190 drugs for high blood pressure, diabetes, and depression.
Here’s the catch: the AI only handles renewals for meds a licensed provider already prescribed. It does not make new treatment decisions.
To qualify, a patient confirms they are physically in Utah. They answer clinical questions. If the system clears them, the refill goes straight to the pharmacy. If the AI hesitates? It hands the case off to a human doctor on Doctronic’s telehealth team.
“We were essentially told: ‘Yes this is going on. And yes, you don’t have a say in it.’” — Dr. Alan Smith, Utah Medical Licensing Board
It sounds simple. But Stanford’s Michelle Mello calls it “one of the first deployments at scale of any autonomous, agentic system in medicine.” That distinction matters. By July, the Associated Press had already started asking doctors and lawyers if we had accidentally granted a machine a medical license.
Early Data: Caution Over Efficiency
The Utah Department of Commerce released pilot data in May. Five months of results. The findings are nuanced, to say the least.
In 72% of cases where Doctronic recommended a renewal, it routed the request to a human physician for final sign-off. Physicians agreed 91% of the time. For that remaining 9%, they asked for more info. Updated labs. More context.
The other 28% of the time, the AI escalated the case to a physician on its own. It flagged complications. It demanded new testing. Physicians agreed with these escalations 69% of the time. That left a 31% chunk where doctors thought the AI was being overly cautious.
Zach Boyd, heading Utah’s AI office, admits Doctronic errs on the side of caution. It routinely kicks uncontroversial decisions up to humans. This is seen as appropriate for a new system. It is also precisely what undermines the efficiency argument. Why automate work if the human still has to do it 72% of the time?
The design guarantees this overlap. A Utah-licensed clinician must personally review the first 250 individual cases before anything hits a pharmacy. Only after the system clears an accuracy threshold will direct human review drop. For now, the overlap is structural, not accidental.
The Hornet’s Nest of Physician Pushback
The goal is obvious: automate the administrative drag of prescription renewals. Patients call or email their doctor’s office. There is no visit tied to the request, so no reimbursement for the staff handling it. Chronic-disease meds rarely change from refill to refill. Logic suggests AI can absorb that noise.
Dr. Adam Oskowitz, Doctronic’s co-founder, envisions a future where AI handles routine tasks like ordering tests. He wants doctors managing “thousands more patients than they can today.”
The Utah Medical Licensing Board isn’t buying it. They learned about the program from news headlines, not state officials. In April, 11 board members wrote a letter demanding suspension.
Their argument? Renewing medications without checking for changes in a patient’s condition is risky. Some of the drugs on Doctronic’s list include blood thinners.
“Many times when I see people after six monthsl I find that their medical history has changed,” Dr. Smith said. “Just because something was prescribed before doesn’t mean it’s appropriate now.”
The AMA echoed this. They warned that prescription renewals aren’t routine checkboxes. They are medical decisions.
Public Citizen took the fight further. In May, they backed the medical board’s call for suspension. They warned that “physician supervised” pilots risk sliding into rubber-stamp approvals as volume grows.
Two Structural Risks: Accountability and Scope Creep
According to Michelle Mello, there are two glaring issues lurking under the surface.
First, accountability. The contract between Utah and Dothronic is murky. Who pays if an AI-driven error injures a patient? The lines are blurred.
Second, scope creep. Once a vendor’s tool deploys through a regulatory workaround, it tends to expand. It grows beyond its original intent without facing the scrutiny of the initial launch. Mello puts it sharply: “If Utah’s pilot program is the camel’s nose in the tent, it will need someone holding the reins when the camel emerges.”
That camel is already emerging. In March, Utah signed a new agreement with Legion Health. This one covers a defined list of non-controlled psychiatric maintenance drugs—SSRIs, SNRIs. It includes mandatory escalation for suicidality, mania, or severe side effects.
The guardrails are tighter. It requires a 98% physician concordance rate for the first 250 cases. Then 99% for the next 1,0,0.
The Regulatory Gap: Who Is Licenssed Here?
Physicians have licenses to practice medicine. AI does not. This binary is breaking the current legal framework.
Eric Bressman at the University of Pennsyvlania argues we’ve crossed a threshold. We are effectively granting a non-human entity a medical license. He compares today’s patchwork to the pre-national licensing standards era of American medicine. He faults Utah for accepting good-faith assurances from the company rather than demanding published outcomes data before launch.
The only evidence available is a company-authored, non-peer-reviewed studying finding Doctronic’s diagnoses matched human doctors 80% of the time in 500 telehealth consultations.
Then there’s jurisdiction. Medical devices fall under federal FDA regulation. Medical practice is a state-level affair. Doctronic sits in the gap between the two. Executives told the AP they view their tool as part of state-regulated medicine. They declined to confirm if they sought FDA clearance.
The FDA responded simply: they have “not authorized any AI chatbots” and are currently taking a hands-off posture.
Where Does AI Healthcare Regulation Go From Here?
Utah won’t be the outlier for long. Reports suggest Texas and Wyoming are weaving similar AI carve-outs into their rules. Lawmakers in Iowa, Idaho, and elsewhere are introducing bills to formally license AI medical services, many using a template from the Cicero Institute.
Adam Meier, Cicero’s health policy director, frames the resistance in economic terms. Whoever goes first gets shot at. Economic interests. Workforce concerns. Job security.
Utah promises to publish its findings. Doctronic says peer-reviewed studies are coming this year. Whether those studies arrive before or after the program drops its human-review phase will likely decide everything.
Companies can expand their business models by leaping beyond current evidence. It’s profitable in the short term. The risk, as Daniel Aaron told the AP, is compromising public trust. Fueling backlash. Long-term, that might cost more than the efficiency gains were ever worth.
