Building Trustworthy AI in Healthcare

Abstract

Artificial Intelligence (AI) has introduced significant innovations in healthcare, yet its adoption has been slower compared to other industries. A key barrier lies in concerns over trustworthiness, as healthcare stakeholders often express skepticism stemming from the opacity of AI systems, risks of privacy breaches, potential biases that exacerbate social inequities, and unresolved safety issues. Ensuring trustworthy AI is therefore essential for its safe, effective, and responsible integration into clinical practice. This integrative review examines four fundamental dimensions of trustworthy AI in healthcare: transparency and explainability, privacy and security, robustness and safety, and bias and fairness. Unlike prior surveys that either addressed trustworthiness in AI broadly or focused narrowly on traditional machine learning and deep learning (DL), this work extends the discussion to Large Language Models and Agentic AI, both of which are poised to play transformative roles in healthcare. We also explore the intersections among trustworthiness dimensions and highlight the often-overlooked roles of user trust and organizational accountability in shaping real-world adoption. Finally, we provide targeted recommendations, outlining practical tools, frameworks, and strategies tailored to different model classes to address concrete trustworthiness challenges in healthcare AI.