Abstract
Healthcare-AI ethics is often examined through normative principles or research already framed as ethical, overlooking ethically significant findings that emerge in real-world implementation without being explicitly framed as ethical. To address this gap, this study examined which explicit and implicit ethical concepts and considerations emerge when AI is implemented in real-world healthcare, how they are framed, where they are situated, and how they are addressed. From 298 ethically relevant extracts, seven themes and 27 subthemes were developed through a predominantly inductive thematic synthesis and subsequently interpreted through four conceptual dimensions: ethical concepts and considerations, ethical framing, ethical locus, and ethical response. Ethically significant findings concerned trust, professional agency, accountability, clinical fit and safety, equity and patient experience, perceived value, and socio-technical alignment. They were often expressed as implementation concerns rather than explicit ethical claims. Causes, consequences, responsibility, and capacity for action were frequently distributed across technical, clinical, organizational, and patient levels, while responses concentrated at the point of use, shifting compensatory work toward clinicians or patients even when problems originated upstream. The synthesis shows that empirical implementation findings can themselves constitute ethical evidence. Comparison across framing, locus, and response generated the concept of cross-dimensional alignment: whether problem framing reflects relevant conditions, responsibility corresponds to meaningful capacity for action, and responses reach conditions capable of change. Responsible healthcare AI therefore depends not only on identifying what is ethically at stake, but on whether ethical governance has practical traction on the socio-technical conditions through which benefits, burdens, risks, and responsibilities are produced.

This publication has 86 references indexed in Scilit: