Could AI Be Licensed to Practice Oncology?
Is AI poised to practice medicine? It may be already. Earlier this year, the state of Utah allowed Doctronic, a heath technology company using AI to make clinical decisions autonomously, to renew prescriptions for patients who request the service.1 Although Utah’s pilot program focuses on medications for common chronic diseases, including hypertension, depression, and diabetes, there is clear potential for broader application in oncology. Such application could extend beyond prescribing medications to also include composing prior authorization letters and monitoring patients for treatment side effects. This column considers the legal and ethical ramifications of AI-driven prescribing in oncology.
Weighing the Pros and Cons of AI-Driven Prescription Renewals
According to the terms of the agreement, Utah’s program permits AI to prescribe 192 drugs to adult patients. Although none constitute a chemotherapy regimen, a few of the listed medications, such as antinausea drugs and corticosteroids, are commonly used to manage the side effects of chemotherapy and radiation therapy in cancer care.
Under the parameters of the program, physicians hired by Doctronic will review AI’s output for the first 250 patients before the system proceeds with the prescription renewal request, and review the next 1,000 patients retrospectively after the AI agent starts responding autonomously. Medication requests for which physician review is requested, or that meet other criteria, such as drug-related problems or changes in the patient’s health status, are then escalated to physician review.
In exchange for permitting the launch of the pilot program, Utah required Doctronic to adhere to a contract that provides consumer safety and privacy protections, disclosure that patients are interacting with an AI agent, and performance monitoring, including user complaints and physician observations. The contract also requires the company to compensate Utah for any liability costs the state incurs. However, it relieves Doctronic of obligations to “generate, maintain, and make available to each patient” the patient’s medical records, and allows the company to require participants to agree to terms of service that disclaim liability for system errors or harmful outcomes.
Understanding the Implications of Using AI Algorithms in Oncology Care
The Utah program focuses on low-risk, chronic conditions rather than on high-risk conditions, such as cancer. However, because of the high incidence rate of cancer in the United States—over 2 million new cancer diagnoses are expected in 20262—it is likely that some consumers participating in this program have a history of cancer, which may not take into account the complex interaction of oral oncolytic drugs, supportive care medications, and changing cancer status. More generally, there is likely to be interest in expanding the program in the future to cover higher-acuity interventions.
When the Utah program was launched, some medical societies, such as the American Medical Association, cautioned against using AI chatbot prescribing, stating that physicians should “remain at the forefront of decision-making and to validate AI outputs to ensure accuracy and patient safety.”3 Other experts, however, reacted more positively, seeing AI prescribing as a low-risk way to help address disparities in accessing physician care and prescription medication refills, particularly in rural areas where travel and copays for a doctor visit may present a major barrier to care.3
The greater complexity of drug prescribing in oncology makes shifting over to an AI-first approach, like Utah’s, less initially appealing. However, oncologists are already integrating AI into existing practice workflows in which physicians still serve as the final prescribers. For instance, some have proposed using AI algorithms to augment dose selection, escalation, and optimization in future clinical trials.4 These approaches represent physician-complementing, rather than physician-substituting, uses of AI.5
Beyond dosing support, the prior authorization process represents a particularly attractive target for AI assistance in oncology. Oncologists and their staff spend substantial time drafting letters to insurers justifying the use of expensive therapies, and delays in authorization can postpone treatment in clinically meaningful ways. An AI agent capable of assembling the relevant clinical history from electronic health records and matching it against payer criteria, citing applicable guidelines, and producing a coherent appeal could significantly accelerate approvals and meaningfully reduce administrative burden. Unlike medication prescribing, AI-driven prior authorization drafting carries lower direct risk to patients, since a physician would still review and sign the final letter.
The Utah model, in which AI output is initially reviewed by clinicians and later subject to retrospective audit, maps neatly into this use case. The potential to streamline the prior authorization process using AI, however, should not obscure ongoing problems with requiring prior authorization for well-established interventions, which could still involve significant treatment delays, high administrative burden, and coverage denials.
Assigning Liability When AI-Driven Medical Errors Occur
Who bears liability, though, when an AI chatbot fails to escalate a symptom that later proves serious? Utah’s contract resolved this concern by shifting liability onto the vendor while permitting the vendor to disclaim it through terms of service. This arrangement may prove less defensible in the oncology context in which the stakes of error are higher and patients are often acutely vulnerable.
- Govind Persad, JD, PhD
Treatment side-effect monitoring offers another promising AI application in oncology. Patients receiving chemotherapy, immunotherapy, or radiation therapy often experience symptoms between office visits that they may underreport or fail to recognize as urgent. An AI agent that conducts structured check-ins, flags worrisome patterns, and escalates to a human clinician when thresholds are met could function as a kind of always-available triage layer.
Who bears liability, though, when an AI chatbot fails to escalate a symptom that later proves serious? Utah’s contract resolved this concern by shifting liability onto the vendor while permitting the vendor to disclaim it through terms of service. This arrangement may prove less defensible in the oncology context in which the stakes of error are higher and patients are often acutely vulnerable.
Expanding the Utah framework to the oncology setting could also require rethinking patient consent. Patients with chronic diseases seeking renewal of a stable medication face a relatively limited-risk decision. By contrast, patients with cancer confronting treatment-related decisions are often navigating prognostic uncertainty, emotional distress, and complex tradeoffs.
What constitutes meaningful patient authorization for AI involvement in their care is a question regulators will need to address. So is whether AI-specific consent may be unnecessary in certain particular contexts once the use of AI in oncology applications is widespread, particularly when the AI output is part of a workflow that ends with a physician signoff.6
Conclusion
Utah’s pilot program allowing an autonomous AI system to renew specific prescription drugs for chronic conditions offers a template that other states and medical specialties may adapt. The complexity of oncology prescribing makes it unlikely that the field will be an early adopter in which AI assumes a primary prescribing role.
However, the same underlying tools can be deployed in lower-risk supporting roles, including drafting prior authorization letters, monitoring patients between visits, and assisting with dose optimization within physician-led workflows. But these applications raise ethical and legal questions spanning informed consent, liability allocation, recordkeeping, and the proper division of labor between AI and clinicians.
Oncologists, medical societies, and federal regulators would do well to engage with these questions now, while the technology is still being shaped, rather than after it has been deployed at scale.
DISCLOSURE: Dr. Persad receives grant funding from the Greenwall Foundation.
REFERENCES
Mello MM: Utah’s experiment with AI-driven prescription renewals. JAMA Health Forum 7(3):e261001, 2026.
Siegel RL, Kratzer TB, Wagle NS, et al: Cancer statistics, 2026. CA Cancer J Clin 76(1):e70043, 2026.
Wu D: Utah launches first-in-the-nation trial that lets AI renew your prescription. The Washington Post, January 8, 2026. Available www.washingtonpost.com/nation/2026/01/08/ai-prescription-drugs-utah.
Ravishankar A, Sundar R: Optimizing Oncology Drug Dosing: Is Artificial Intelligence the Future? ASCO Daily News, February 21, 2024. Available at https://dailynews.ascopubs.org/do/optimizing-oncology-drug-dosing-artificial-intelligence-future.
Chen J, Gale RP: Physician-Complementing Artificial Intelligence in Oncology. ASCO AI in Oncology, October 29, 2025. Available at https://ascoai.org/articles/2025/physician-complementing-artificial-intelligence-in-oncology/.
Mello MM, Char D, Xu SH: Ethical obligations to inform patients about the use of AI tools. JAMA 334(9):767-770, 2025.
Dr. Persad is Assistant Professor at the University of Colorado Law School.
Editor’s Note: This column is meant to provide general information about legal topics, not legal advice. The law is complex, varying from state to state, and each factual situation is different. Readers are advised to seek advice from their own attorney.
Disclaimer: This commentary represents the views of the author and may not necessarily reflect the views of ASCO, Conexiant, or ASCO AI in Oncology.