An international consortium developed prognostic models for anal cancer using federated learning to train algorithms across multiple centers without sharing raw patient data. The work, detailed in a forthcoming Nature Communications paper, produced models that generalize across diverse cohorts while preserving data privacy and institutional control. Researchers report improved risk stratification compared with single‑center models and stress federated learning’s value for rare cancers where pooled data transfer is logistically and legally challenging. The study incorporated clinical, imaging and molecular inputs across sites. Authors propose prospective validation and clinical‑decision integration to assess whether federated prognostics can guide treatment selection and trial enrollment for anal cancer.