An international consortium used federated learning to develop prognostic models for anal cancer, training algorithms across multiple centers without sharing patient-level data. The Nature Communications paper described model development, external validation and how federated approaches preserved data privacy while delivering generalizable prognostic performance across diverse cohorts. Teams reported that federated models matched or outperformed centrally trained models in several metrics, demonstrating robustness across imaging and clinical heterogeneity. The study highlighted model interpretability efforts and proposed clinical risk stratification use cases for treatment planning and trial enrichment. Authors emphasized federated learning as a path to multi-institutional AI in oncology where data governance or regulation prevents centralized pooling. The approach reduces legal and logistical barriers and supports collaborative model improvement while retaining local control of patient data. AI vendors, hospital IT groups and clinical trial designers should consider federated pipelines for multi-site model development. Regulatory reviewers will need documentation standards for federated model training, validation, and continuous monitoring in clinical deployment.