This review evaluates the generalizability of AI pathology models in predicting EGFR mutations, highlighting significant performance variations across genetic ancestries and specimen types, and their role in clinical triage.