Operationalising Trust: From Human Participation to Living Benchmarks in Healthcare AI

The growing integration of Artificial Intelligence (AI) into clinical decision-making has heightened the concerns regarding its trustworthy deployment in healthcare. Although there is broad agreement on ethical principles of trustworthy AI, including privacy, fairness, robustness, there remains a lack of consensus on how these principles should be operationalised and translated into practice within real-world clinical settings. Drawing on the discussions from a one-day interdisciplinary workshop, this perspective identifies three interconnected system-level components to address this gap: (1) human-centred infrastructures that enable participation with privacy as a sustained precondition; (2) explainability and uncertainty quantification mechanisms, tailored to context, that support accountability and actionability; and (3) evaluation that evolve with both shifting models and shifting standards that evaluate them. We argue that these components should be mutually reinforcing, otherwise advances in each remain local and do not translate into a system-level trust. We present actor-specific actions and governance questions as a starting point for community deliberation and collective action on the operationalisation of trustworthy AI in healthcare. Preprint status: This manuscript is made available as a preprint. TrustworthyAI4Health Workshop 2026: https://elsa-ai.eu/elsa-workshop-trustworthyai4health/