Artificial Intelligence Can Now Prescribe Medications. Utah Opens the Doors to 'Algorithmic Doctors'

We are used to seeing artificial intelligence optimize workflows, generate images, or write code. However, the boundary has now been crossed into the delicate field of public health. Utah has officially become the first American state to authorize an AI system to prescribe medications autonomously.

The Regulatory Sandbox: Where Innovation Moves Fast

The initiative is under the auspices of the Utah Office of Artificial Intelligence Policy, an entity created in 2024 with the aim of making the state a hub for technological innovation. Thanks to a "regulatory sandbox" framework—a sort of regulatory free zone—the startup Doctronic has obtained approval for a 12-month pilot program.

The system is not authorized to issue new diagnoses but can manage the renewal of approximately 190 medications for chronic conditions. For obvious safety reasons, opioids, injectable medications, and treatments for ADHD remain excluded. The stated goal is to reduce costs ($4 per renewal) and tackle the issue of therapeutic non-adherence, which causes about 125,000 preventable deaths a year in the USA due to cumbersome bureaucratic processes.

Doctronic, founded by vascular surgeon Adam Oskowitz and CEO Matt Pavelle, is not a newcomer. The company has raised over $65 million in funding and boasts a dataset of 15 million consultations. According to data provided to regulators, the AI has shown a 99.2% alignment with the decisions made by human doctors on a sample of 500 cases. "We are not only faster, we are more meticulous," say the founders. The algorithm can perform dozens of cross-checks on safety and drug interactions in a matter of seconds, a task that would take a human clinician an incompatible amount of time with the pace of modern medicine.

Cybersecurity: The Achilles' Heel

Despite the technological enthusiasm, shadows loom. Recently, the security company Mindgard demonstrated how a public chatbot from Doctronic could be manipulated through "prompt injection" attacks. In a test, researchers convinced the AI to triple a dose of OxyContin (a powerful painkiller) using a fake regulatory bulletin. Although Doctronic clarified that the system used in the pilot program is isolated and much more protected than the public chatbot, the case raised significant concerns. If a language model can be "tricked" by malicious inputs, how safe is it to entrust the health of thousands of citizens to it?

Reactions from the medical community have been swift. The American Medical Association (AMA) has expressed strong reservations, emphasizing that the absence of human clinical judgment poses incalculable risks for patients. The main criticism concerns oversight: if in time, the AI will autonomously manage 95% of renewals with only sampling checks, who will be responsible in case of a fatal error?

As Utah blazes the trail, other states like Arizona and Texas are watching closely, ready to launch their own sandboxes. If the 12-month test yields positive results, we could be facing a structural change in global healthcare, where the physician is no longer the terminal for prescriptions, but the supervisor of an increasingly pervasive algorithmic infrastructure.