Who is Accountable When Clinical AI Fails? A Governance Framework for Health Systems in Africa

Artificial intelligence is being deployed in clinical settings across Africa faster than health systems are building the governance structures to manage it. The result is a dangerous accountability vacuum: when AI-assisted clinical decisions cause harm, responsibility is diffused across clinicians, administrators, vendors, and IT departments until it effectively disappears. Existing health governance frameworks in African countries-including Nigeria and South Africa-were designed for human clinical decision-making and do not anticipate algorithmic decision-support. Global frameworks such as the EU AI Act (2024), the African Union AI Initiative (2024), and UNESCO's Recommendation on AI Ethics (2021) provide important principles but do not resolve the practical question of who is accountable in a specific clinical setting when a specific AI system fails. This paper proposes a practical Clinical AI Accountability Framework comprising four named governance roles-the Clinical Decision Owner, the Institutional Governance Lead, the Technical Accountability Officer, and the Patient Advocacy Interface-operating across three AI lifecycle stages: predeployment, active deployment, and post-incident response. The framework is operationalised through a five-question Clinical AI Governance Audit. The paper argues that accountability in clinical AI does not require new legislation-it requires named responsibility, documented roles, and institutional commitment to implementation.