Utah lets Legion Health's chatbot renew psychiatric prescriptions

Utah has licensed a chatbot from the startup Legion Health to renew prescriptions for psychiatric drugs, with no mandatory physician involvement. It is only the second time in US history that clinical authority has been formally handed to an artificial intelligence system.

The one-year pilot, announced last week, works in a simple way: patients in Utah pay $19 a month for a subscription, talk to the Legion Health chatbot and, in some cases, receive renewed prescriptions for psychiatric medication without consulting a live psychiatrist or primary care doctor. The San Francisco company pitches the product as a "fast and easy" way to extend prescriptions.

Utah officials name two main goals for the experiment: lowering the cost of psychiatric care and partly closing an acute shortage of specialists. The US has a severe shortage of psychiatrists: the Association of American Medical Colleges estimates that by 2036 the deficit of mental health physicians could reach 30,000. Against that backdrop, authorities see delegating routine tasks to an algorithm as a reasonable compromise.

The medical community is skeptical. Doctors point to several problems. First, the chatbot is a black box: it is unknown what data and logic it uses to make clinical decisions, or how it tracks side effects and drug interactions. Second, psychiatric drugs are not ordinary medications: antidepressants, antipsychotics and mood stabilizers require regular, live assessment of the patient's condition. Third, critics argue the $19 subscription model is more likely to serve people who already have access to healthcare than to help low-income patients, the uninsured and residents of remote areas.

It is important to understand the scale: this is about renewing existing prescriptions in select cases, not making an initial diagnosis. Still, it is a clear shift in regulatory logic: US states are increasingly willing to test AI in clinical practice without waiting for federal standards.

The key question remains unanswered: who is liable if the chatbot extends a prescription for a patient for whom it is contraindicated? What is the escalation procedure if the system "sees" alarming symptoms? Legion Health does not disclose details of its architecture or safety mechanisms.

If the one-year pilot ends without serious incidents, it will be a weighty argument for regulators across the country. Psychiatry is especially sensitive to errors: the cost of a wrong decision is measured not in discomfort but in crises and hospitalizations. Handing this field to AI without strict transparency standards means building trust on an advance the system has not yet earned.