JMIR Publications

ABSTRACT

Large language models (LLMs) are increasingly integrated into everyday health-care communication, moving beyond experimental evaluation into routine clinical and informational use. Early research has primarily focused on technical performance, including accuracy, validation, and bias mitigation. While these remain essential, the transition to sustained real-world integration raises additional ethical questions that cannot be addressed by model-level evaluation alone. This Viewpoint proposes an adoption-phase ethics perspective, emphasizing how ethical risks shift as LLMs become embedded within institutional workflows, professional practices, and relationships of care. Drawing on normative analysis informed by existing empirical discussions, we examine three interrelated domains: trust, responsibility, and equity. During routine use, trust becomes shaped not only by perceptions of accuracy but also by expectations regarding accountability, transparency, and institutional protection. Responsibility may become diffused or ambiguous when LLM-mediated information influences clinical communication without clearly specified oversight. At the same time, differential digital literacy and access to institutional support may create uneven capacity to interpret and benefit from AI-generated information. We argue that ethical governance must therefore extend beyond pre-deployment technical safeguards toward sustained, system-level oversight. Adoption should be understood as a dynamic ethical process requiring role-sensitive design, clear accountability structures, and equity-oriented implementation. By reframing ethical attention from experimental validation to governance during routine integration, health-care systems can better ensure that the growing presence of LLMs supports fairness, responsibility, and patient trust alongside technical advancement.