Abstract Pathology foundation models (PFMs) are large-scale pretrained models tailored for computational pathology, substantially advancing the development of downstream models across a wide array of diagnostic tasks. However, their clinical deployment raises ethical concerns, including privacy leakage, reliability in diverse clinical settings, and fairness across patient subgroups. These risks remain underexplored and insufficiently quantified, yet addressing them is critical to enable the safe translation of PFMs from research to clinical practice. In this study, we propose the first quantitative framework to systematically assess ethical risks in PFMs. We focus on three critical dimensions: potential leakage of patient-sensitive attributes, reliance on non-diagnostic features that compromise reliability, and subgroup disparities in performance. To this end, we design task-specific experiments across a wide range of datasets and clinical scenarios, providing a quantitative and auditable protocol for ethical risk evaluation. The results show that while current PFMs demonstrate strong performance, they still exhibit challenges in patient privacy, clinical reliability, and group fairness. We further explore the underlying causes of these ethical risks in PFMs, substantiate our findings with empirical evidence, and provide insights into strategies for their mitigation. This work provides the first quantitative and systematic assessment of ethical risks in PFMs. Our findings highlight the urgent need for ethical safeguards in PFMs and offer actionable insights for building more trustworthy and clinically robust PFMs. We will release these evaluation results and provide the assessment framework as a toolkit to support the development, auditing, and deployment of PFMs in both research and translational settings.
Wang et al. (Wed,) studied this question.