Fairness in health AI isn't just a model performance problem; it's a lifecycle challenge. In a new guest editorial published in the American Journal of Bioethics, Duke Health AI Evaluation & Governance Program’s Matthew Elmore, Nicoleta Economou, and colleagues discuss how bias can emerge through data, workflows, implementation decisions, human-AI interaction, and ongoing monitoring. The authors offer practical recommendations for addressing those risks across the AI lifecycle, recognizing that responsible AI requires attention at every stage, from development and evaluation to implementation and continuous monitoring.