Beyond generalist AI: purpose-built foundation models for transplantation

Solid organ transplantation is a defining test case for clinically grounded foundation models. These models have begun to transform language, imaging, and pathology elsewhere in medicine, yet many remain built around a single patient and a static snapshot. Transplant outcomes instead arise from a dynamic donor-recipient-graft ecosystem across the care continuum, under organ scarcity and fragmented multimodal data. Published work has adapted generalist models to isolated tasks, showing component feasibility rather than transplant-native pretraining. In this Perspective, we explain why generalist medical AI may leave transplant care's needs unmet. We define five requirements for transplant-native models: joint donor-recipient-graft representation, longitudinal continuum memory, multimodal integration, decision awareness, and continual adaptation and grounding under human oversight. We then set out a staged roadmap from data infrastructure and model development through clinical validation to governed, assistive deployment. Early use should remain bounded and assistive, judged by compatibility with transplant biology, clinical workflow, patient safety, and ethical allocation rather than scale alone.