Utah will start a one-year pilot this month in which artificial intelligence renews prescriptions for some psychiatric drugs without a doctor visit. It is limited to 15 low-risk maintenance medicines that have already been prescribed and to patients in stable condition.

The Verge reported on April 3 that the project will be handled by an AI chatbot run by telemedicine startup Region Health. It is the second case in Utah and the United States in which such clinical authority has been given to AI. The company is offering the service as a $19-a-month subscription and is currently taking names for a waiting list.

The chatbot cannot issue new prescriptions. It can renew only medicines already prescribed by a clinician, including fluoxetine, sertraline, bupropion, mirtazapine and hydroxyzine. Patients who have recently changed dosage or medication, or who have been hospitalised for psychiatric reasons within the past year, are excluded. Patients must meet medical staff once every 10 refills or once every six months.

Also excluded are medicines that require blood-test monitoring, controlled drugs, benzodiazepines, antipsychotics and lithium. Many ADHD treatments and complex cases such as schizophrenia and bipolar disorder are outside the scope of this pilot.

Users must consent to participate, then verify their identity and prove their existing prescriptions. The chatbot asks about symptoms, effectiveness, side effects, suicidal thoughts, self-harm, severe reactions and pregnancy status. If a case falls outside the low-risk criteria, a clinician reviews it again. Patients and pharmacists can also request human review.

Psychiatrists said the service is unlikely to significantly expand access because it can be used only by patients already in a treatment plan. Brent Kious (브렌트 키우스), a professor at the University of Utah School of Medicine, and John Torous (존 토러스), a professor at Harvard Medical School, said a chatbot could miss a patient’s condition during decisions on maintaining, tapering or stopping medication and during the questioning process.

Utah has required doctors to closely review the first 1,250 cases and to conduct spot checks of about 5 to 10 percent thereafter. In another AI prescribing pilot that began in December, cases included spreading vaccine conspiracy theories, generating instructions for making methamphetamine and tripling an opioid dose.

The pilot reflects practical needs to expand access to care and ease staffing shortages, but whether psychiatric care can be standardised remains a separate question. Even with the scope narrowed to low-risk patients and limited drugs, it has yet to be verified whether AI can properly screen subtle nuances in questioning and changes in condition. In that sense, the pilot is seen as testing the boundaries of safety and accountability more than efficiency.