This article contends that the European frameworks governing clinical AI fail to sustain accountability once systems are in use, and that certification and documentation are not apt to capture distributed, evolving decision-making. Drawing on comparative doctrine, the paper diagnoses the gap between formal compliance and lived responsibility. It advances living oversight as a governance practice that is continuous and substantively human, reframes stewardship as a shared responsibility across developers, clinicians, institutions, and regulators, and sets out practical implications: dynamic consent, post-deployment auditing, and enforceable redress. Accountability has to be designed into operation – not verified after the event – if clinical AI is to remain compatible with the moral foundations of healthcare.
COSTA et al. (Mon,) studied this question.