Governing AI in Radiotherapy: Ensuring Automation Supports Rather Than Substitutes for Workforce Expertise

Radiotherapy (RT) is required for approximately half of all patients with cancer, yet global access remains profoundly unequal, particularly in low- and middle-income countries (LMICs), where workforce growth has failed to keep pace with rising cancer incidence, and most who require RT do not receive it. Artificial intelligence (AI) is increasingly framed as a solution to this gap, with automation promoted to "leapfrog" workforce shortages by accelerating contouring, treatment planning, and quality assurance. We argue that explicit governance and workforce safeguards are needed to complement AI proliferation, to mitigate risks of overreliance on automation and reduced opportunities for clinical expertise development. Diffusion optimism (the assumption that introducing AI into RT workflows will automatically yield safe, scalable clinical capacity) and its companion, validation optimism (the assumption that regulatory clearance implies safety across diverse clinical contexts) rest on the premise that deployment will generate efficient clinical capacity by default. In RT, this premise is difficult to separate from the formation and preservation of expertise. Without governance, accountability, and training safeguards, AI not only introduces technical risk but reshapes clinical authority in ways that can perpetuate algorithmic dependency and reduce opportunities to develop clinical expertise. We therefore argue that AI must be governed as a capacity-shaping technology, prioritizing how expertise is formed, exercised, and preserved, rather than as an outright capacity substitute.