A new review in the Journal of Translational Medicine argues that AI-derived biomarkers could improve prediction of cancer immunotherapy response where markers like PD-L1, tumor mutational burden, and microsatellite instability remain imperfect. It highlights deep learning approaches in digital pathology and radiomics that quantify tumor-immune features and spatial immune phenotypes. The review also catalogs AI work in resistance prediction to immune checkpoint inhibitors, immunotherapy candidate stratification, and forecasting immune-related adverse events. It stresses caution around model interpretability and variability across cancer types, endpoints, and validation designs. Overall, it frames AI as a technically capable but still method-sensitive toolkit that will require rigorous, consistent validation to support clinical adoption.