From principle to practice: Developing a digital-age clinical artificial intelligence ethics competence framework...

As artificial intelligence (AI) becomes increasingly embedded in clinical workflows, clinicians encounter ethical challenges that traditional, principle-based medical ethics education may not adequately address. Empirical evidence on clinicians' experiences with AI-related ethics is limited, constraining curricular improvement. This qualitative study explores how early-career doctors in Singapore perceive and negotiate ethical dilemmas arising from clinical AI use and translates findings into an operationalised competence framework for medical education.

Between April and June 2025, we conducted semi-structured interviews with 30 early-career doctors (1-5 years post-graduation) from nine public healthcare institutions in Singapore. Purposive sampling ensured diversity across specialties, institutions, gender and ethnicity. Interviews explored participants' AI-related ethical challenges in day-to-day practice and their perceptions of ethics training in medical school. Data were analysed using Braun and Clarke's (2022) reflexive thematic analysis, with codes developed iteratively and informed by the four classical bioethical principles as sensitising concepts-autonomy, beneficence, non-maleficence and justice. Interdisciplinary reflexive discussions guided the construction and interpretation of themes.

Participants reported limited formal AI education. Seven recurring practical ethical challenges were identified: (1) system opacity, (2) dataset bias and generalisability, (3) data privacy and consent in networked environments, (4) insufficient patient-specific contextualisation of outputs, (5) risks of hallucinations, (6) ambiguous accountability and (7) cognitive offloading. These themes reframed classical bioethical principles through epistemic, relational and institutional lenses.

Ethical competence for AI-mediated care requires integrated epistemic and relational capacities beyond technical literacy or traditional medical ethics. We propose the Digital-Age Clinical AI Ethics Competence (DCEC) framework, comprising four domains of epistemic awareness, relational integrity, reflexive accountability and adaptive professionalism, anchored by ethical digital literacy (EDL). Each domain is operationalised with specific learning activities and assessment strategies such as Objective Structured Clinical Examination (OSCE) stations, reflective portfolios and ethics viva. We discuss implications for curriculum design, faculty development and competency-based assessment.