In the high desert capital of Salt Lake City, a quiet experiment has upended decades of medical tradition. Since January, Utah residents with chronic conditions no longer need to schedule a doctor’s appointment for many routine prescription renewals. They chat with an AI.
The system, built by New York-based startup Doctronic, reviews patient history, verifies identity through selfies and ID photos, asks about symptoms or changes in health, and then decides. Approved refills flow straight to a local pharmacy. The cost? Just $4 per renewal. The state calls it a first-of-its-kind test of autonomous artificial intelligence in clinical decisions. Utah’s Department of Commerce announcement framed the partnership as a way to close gaps in medication adherence that contribute to poor health outcomes and higher costs.
But not everyone cheers the development. Eleven members of the Utah Medical Board fired off a letter in April demanding an immediate halt. They argued the AI could miss drug interactions, overlook side effects or fail to catch shifts in a patient’s condition. Blood thinners drew special worry. “Overseeing prescription refills is a task reserved for licensed physicians,” the board stated, according to reports in The Next Web.
The state pushed back. Officials told the board it had “no say” in the matter. The pilot operates inside Utah’s regulatory sandbox, created under 2024 legislation that lets the Office of Artificial Intelligence Policy waive certain licensing rules for tested technologies. Doctronic’s AI handles renewals for about 190 medications used in chronic care. Exclusions cover controlled substances, injectables, painkillers and ADHD treatments. Initial prescriptions still require a human doctor. The AI only steps in afterward.
Here’s how it works in practice. A patient logs on, confirms location in Utah, uploads verification and selects from their existing prescriptions. The chatbot probes for updates. If everything checks out against guidelines and history, it generates the refill order. Complex cases escalate to a physician on Doctronic’s staff. Early on, doctors reviewed the first 250 approvals in each category. Later reviews became retrospective on a sampling basis.
Proponents see clear upside. Patients avoid weeks-long waits for appointments. Pharmacies process orders faster. Physicians escape administrative drudgery that crowds out time for complex cases. “Rather than wait days or weeks to get an appointment with their doctor to get a routine medication refilled, Utahns can chat with an AI agent online to renew their prescriptions nearly instantly,” Fierce Healthcare reported in its coverage of the launch.
Yet the speed raises questions that linger. What happens when the AI encounters incomplete data? Or when a patient’s health has changed in ways not captured by self-reporting? Researchers outside the program have already demonstrated vulnerabilities. In March, security firm Mindgard used simple jailbreaking prompts to trick the system into suggesting tripled doses of OxyContin, mislabeling methamphetamine and pushing false claims about vaccines. Those findings, shared first with Axios, cast doubt on safeguards.
Doctronic maintains its system meets the same liability standards as a human doctor through malpractice coverage. It submits monthly reports to state overseers tracking approvals, denials, sampled cases and complaints. The company also hired physicians to validate early outputs. Still, the oversight board for the sandbox consists of AI policy specialists rather than practicing clinicians. That structure troubles many in medicine.
Dr. Alan Smith, chair of the Utah Medical Board, voiced the disconnect. The board learned of the program only after it began. “They told us we had no say,” he said in accounts that spread across outlets including AP News, which published a detailed examination of the debate on July 6. The story captured fresh pushback from physicians wary of handing clinical judgment to algorithms.
Dr. Eric Bressman, a University of Pennsylvania researcher focused on AI in health care, set a high bar. “The threshold for allowing AI to practice medicine independently should be extremely high,” he told AP. His caution reflects wider sentiment. The American Medical Association has echoed concerns about safety, oversight and the potential erosion of trust when machines replace human interaction in sensitive areas.
And the federal government? The Food and Drug Administration has not cleared any AI chatbots for prescribing or renewing medications. Agency officials maintain a hands-off posture for now, leaving states to chart their own course. That vacuum amplifies the stakes in Utah. What starts as a 12-month pilot could influence policy elsewhere. Texas and Wyoming have signaled interest in similar models.
Academic observers offer mixed views. A perspective in the New England Journal of Medicine from April, authored by Sara Gerke, Ravi B. Parikh and I. Glenn Cohen, flags both promise and peril. The program raises “important clinical and legal issues” around autonomous systems in patient care. It could improve access for millions struggling with chronic disease management. Yet liability, accountability and error correction remain unsettled.
JAMA Health Forum weighed in too. Its analysis noted the pilot’s rapid move toward autonomy after initial physician checks on the first 250 patients and retrospective review of the next 1,000. “Although physicians will initially validate the AI’s actions, the pilot program will swiftly become one of the first deployments at scale of an autonomous, agentic system in medicine,” the authors wrote.
Real-world results stay limited so far. The state has released no comprehensive usage statistics. Doctronic reports the system processes renewals for common drugs such as statins, blood pressure medications, certain psychiatric treatments and birth control. Monthly disclosures cover acceptance rates and complaints, but independent verification is scarce. A company-funded study is expected later this year. Skeptics question its objectivity.
Critics also point to economic undercurrents. Routine renewals make up roughly 80 percent of prescription activity in many practices. Shifting that work to AI could reshape physician workloads, clinic revenue and even job prospects for support staff. Some fear it accelerates a broader trend of technology displacing human judgment without sufficient proof of superiority.
Supporters counter with data on nonadherence. Millions of Americans with hypertension, diabetes or heart disease lapse on medications because refills prove too cumbersome. A faster process could yield measurable gains in outcomes and lower hospitalization rates. Utah officials emphasize the pilot’s narrow scope. It never replaces initial diagnoses or high-risk decisions. And escalation protocols exist for uncertainty.
But protocols on paper differ from performance in messy reality. Patient self-reports can be inaccurate. Medical records may contain gaps. Algorithms trained on historical data sometimes inherit biases or fail in edge cases. The Mindgard red-teaming exercise exposed exactly those weaknesses. Simple adversarial prompts bypassed intended guardrails. If malicious actors or even confused users can manipulate outputs, the risks multiply.
So. The program continues. Utah’s AI office defends its cautious design and commitment to evaluation. Doctronic touts transparency and plans for peer-reviewed publication. Physicians on the medical board remain unconvinced. Their April letter called the initiative premature and potentially dangerous. The standoff highlights a deeper tension. Innovation races ahead. Regulation and professional norms trail behind.
Recent coverage underscores the divide. AP News captured doctors’ wariness in its July 6 report, noting the program has “kicked off a big medical debate” about whether AI stands ready for tasks long reserved for physicians. Similar stories appeared across regional outlets, many syndicating the AP piece. On X, clinicians and policy watchers traded sharp takes. One radiologist posted the AP link with a simple warning: the move sets a “precedent-shattering” example.
Others see opportunity. Health-tech advocates argue sandboxes like Utah’s offer controlled environments to gather evidence before wider adoption. Without such tests, the U.S. risks falling behind nations experimenting more aggressively with AI in care delivery. The key lies in rigorous monitoring, independent audits and clear escalation paths. Those elements exist in the Doctronic agreement on paper.
Implementation will decide success or failure. If the pilot demonstrates high accuracy, few adverse events and strong patient satisfaction, other states may follow. If errors surface or public confidence erodes, the backlash could slow AI’s foothold in clinical practice for years. For now, Utah serves as the nation’s test track. Thousands of refills have likely moved through the system already. Yet the medical community watches closely, prescriptions in hand and questions unresolved.
The experiment continues through the end of its 12-month window. Data will flow. Studies will appear. And the conversation about where machines belong in the exam room shows no sign of fading.
WebProNews is an iEntry Publication