There’s A Long Way to Go Before AI Tools Can Prescribe Medication. But How Long?
By Christopher Cheney
Legislation that would allow AI tools to prescribe medication is premature, according to an AI expert at Sutter Health.
AI is taking on increasing roles in clinical care. For example, there are AI tools that help radiologists to read and interpret medical images. In addition, AI tools can provide clinical decision support. Ambient listening AI tools record interactions between clinicians and patients, then develop a clinical note for the electronic health record.
In January, U.S. Rep. David Schweikert, R-Arizona, introduced the Healthy Technology Act of 2025. The legislation, which is being considered by the U.S. House of Representatives Committee on Energy and Commerce, would allow AI tools or machine learning technologies to prescribe medication.
Under the Healthy Technology Act of 2025, an AI tool or machine learning technologies would be considered a practitioner licensed by law to prescribe medication under two circumstances. First, the technology would have to be approved as a practitioner to prescribe medication by state law. Second, the technology would have to be approved, cleared, or authorized by the U.S. Food and Drug Administration.
AI tools or machine learning technologies are not ready to play the role of a practitioner who prescribes medication, according to Kiran Mysore, MS, chief data and analytics officer at Sutter Health, and a HealthLeaders AI in Clinical Care Mastermind member.
“AI exists to supplement physicians, not supplant them,” Mysore says. “AI helps elevate the level of patient care and reduce physicians’ cognitive burden. Use of AI in this fashion is still premature; there are several risks and unknowns that we should work through.”
AI tools can help make decisions on prescribing medications, but a clinician still needs to have the final say, Mysore explains.
“Our general rule of thumb is that we need a human in the loop across the board as a requirement in clinical decision-making,” Mysore says.
Several hurdles must be cleared before AI tools are ready to prescribe medication without the supervision of a clinician, according to Mysore. First and foremost, AI tools must have reliable knowledge and understanding of a wide range of illnesses and treatments.
“It has to have a deep awareness of the patient’s entire medical history and be able to get the latest information easily,” Mysore says. “The tool has to also have internal guardrails to prevent errors and bias, with some level of explainability and chain of reasoning.”
In addition, there needs to be a mechanism to rate actual prescriptions against a physician’s recommendations, Mysore explains.
Christopher Cheney is the CMO editor at HealthLeaders.