Explainable Agentic Artificial Intelligence in Healthcare: A Scoping Review

This scoping review aims to systematically map and characterize the emerging field of explainable agentic artificial intelligence (XAAI) in healthcare. Recent advances in large language models and multi-agent systems have enabled AI systems capable of autonomous, multi-step reasoning, task decomposition, and tool use. While these agentic systems show promise in clinical decision support, diagnostics, and biomedical research, their increasing autonomy raises important concerns regarding transparency, safety, and trustworthiness. At the same time, traditional explainable AI (XAI) research has largely focused on static predictive models rather than dynamic, agentic workflows. The purpose of this review is to identify healthcare-oriented agentic AI systems that incorporate explicit explainability mechanisms and to classify the explainability approaches they use. Specifically, we analyze explainability according to type (intrinsic, post-hoc, hybrid), scope (local, global, or both), and form (textual, feature- or concept-based, and example-based). We also examine how agentic characteristics—such as autonomy, multi-agent architectures, task orchestration, and tool use—relate to the types of explanations generated. We expect this review to clarify current design patterns in XAAI, identify methodological gaps (particularly regarding global explainability and evaluation of explanation fidelity), and provide guidance for future development of safe, transparent, and clinically trustworthy agentic AI systems in healthcare.