Between map and maze: reframing trust in healthcare AI

Abstract Artificial intelligence (AI) is often presented as a transformative technology for healthcare, promising to augment clinical decision-making, streamline workflows, and enhance diagnostic precision. Yet its integration into healthcare practice is shaped by the complex and often ambiguous notion of trust . While trust in AI has become a recurring theme across disciplines, there has been little systematic analysis of different conceptualizations of the term. This paper addresses this gap through an interdisciplinary scoping review that examines how trust, trustworthiness, distrust, and mistrust are articulated in literature on AI in healthcare. Drawing on 82 publications (2015–2025) retrieved from six databases (Web of Science, Scopus, PubMed, PhilPapers, SocINDEX, and ACM Digital Library), it maps how trust is defined, measured, or problematized, and identifies the conceptual, epistemological, and disciplinary boundaries that shape these approaches. The analysis puts forward five different conceptualizations: trust as a set of designable principles , trust as an attitude or belief , trust as a binary between cognition/affect or human/technology, trust as a structural mechanism within uncertainty, and trust as a relational process embedded in socio-technical systems. These framings illuminate how trust is treated as an attribute to be engineered, a behavior to be calibrated, or a relation to be cultivated—often removed from clinical practice. Yet, across literature, trust disruptions which manifest as distrust, mistrust, overtrust, and undertrust are undertheorized and are typically framed as obstacles to overcome rather than phenomena warranting analysis in their own right. By introducing these diverging conceptualizations, the paper argues that trust in healthcare AI is less a stable condition than a dynamic negotiation that reveals power relations, uncertainties, and institutional dependencies of AI. Recognizing distrust and mistrust as analytically productive allows for a more reflexive understanding of how trust is enacted, distributed, and contested within healthcare. This review contributes to an emerging body of work that reframes trust not as a prerequisite for technological adoption but as a contested and dynamic concept.