Ethical Governance of Large Language Models in Health Care: Trust, Responsibility, and Equity in Routine Use.

Abstract

Large language models (LLMs) are becoming increasingly embedded in routine health care communication, raising ethical challenges that extend beyond model performance alone. This Viewpoint argues that ethical risks in LLM-enabled health care emerge through patterns of reliance, institutional embedding, and governance during real-world use. Using "adoption-phase ethics" as an analytic lens, this paper examines 3 interrelated dimensions of ethical risk. First, trust in LLM-enabled health care is shaped not only by technical accuracy, but also by institutional and relational conditions surrounding its use. Second, responsibility may become distributed and ambiguous when LLM-mediated information influences clinical communication and decision-making. Third, equity concerns arise from unequal capacities to interpret, contest, and benefit from LLM-generated information. We argue that ethical governance of LLMs in health care requires continuous, system-level oversight that extends beyond model evaluation alone, including clear accountability structures, role-sensitive implementation, and equity-oriented governance practices. By reframing ethical analysis around routine integration rather than technical performance alone, this Viewpoint aims to support more responsible and sustainable use of LLMs in health care.