From principles to practice: operationalizing responsible AI governance across predictive and generative clinical AI

Responsible artificial intelligence (AI) governance in healthcare is increasingly supported by international principles addressing transparency, accountability, safety, explainability, and lifecycle risk management. Yet a persistent implementation gap remains between these institution-level principles and the concrete controls encountered when an AI service is used in clinical practice. This Perspective argues that responsible AI should be operationalized at the functional points where clinical AI outputs are produced, explained, and communicated. We use three functional domains—Capability, Explainability, and Communication—as an organizing structure for translating existing governance principles into application-level specifications. The approach accommodates both predictive AI (PAI) and generative AI (GAI) while recognizing their different risk profiles. For PAI, governance emphasizes target and prediction-horizon definition, calibration, thresholds, drift, external validation, and alert burden. For GAI, it emphasizes task boundaries, source traceability, hallucination control, version management, human verification, and constraints on write-back or autonomous action. Two dialysis-related applications illustrate how the same governance vocabulary can be instantiated differently across AI modalities. Operationalizing responsible AI in this way does not introduce new governance principles; rather, it organizes existing requirements into observable, reviewable, and auditable controls embedded in clinical AI services and workflows.