The digitisation of histopathology has accelerated the application of artificial intelligence (AI) to cancer diagnosis and precision oncology; however, most deployed AI systems remain narrowly task-specific and difficult to translate across diverse clinical environments. Pathology foundation AI models are emerging as a unifying paradigm, enabling the learning of generalisable representations of tissue morphology through large-scale pre-training and supporting a broad range of downstream tasks. In this narrative review, we examine the development, methodological foundations and current landscape of pathology foundation models in oncological pathology. We outline the evolution and principal trends in the field, classify the major model types and modalities and evaluate their capabilities and advantages in comparison with conventional pathology AI systems. We also examine the transition from foundation models to agentic AI systems and its implications for integrated, workflow-aware pathology practice. In addition, we review relevant regulatory and governance frameworks, with particular attention to requirements for validation, accountability, transparency and oversight.
Sali et al. (Fri,) studied this question.