A review in the Journal of Translational Medicine argues that AI-derived biomarkers could improve prediction of which patients will respond to cancer immunotherapy by integrating multiple data modalities beyond single markers. It cites digital pathology and radiomics methods that quantify tumor-infiltrating lymphocytes and spatial immune phenotypes, as well as multi-omics models such as graph neural networks and transformers. The report highlights applications in predicting primary and acquired checkpoint inhibitor resistance, identifying immunotherapy candidates, and forecasting immune-related adverse events. Importantly, it cautions that explainability outputs should not be treated as causal mechanisms. From a product perspective, this frames the near-term value proposition for AI biomarkers: better discrimination at decision points, but only if models are validated across cohorts with consistent endpoints and clinically meaningful thresholds.