From Principles to Practice: Why Global AI Governance Frameworks Fail at the Point of Clinical Decision-Making and What Africa Needs Instead

The past decade has produced an unprecedented volume of global AI governance frameworks — from the OECD AI Principles (2019) and UNESCO's Recommendation on the Ethics of AI (2021) to the EU AI Act (2024), the Council of Europe AI Treaty (2024), and the African Union AI Initiative (2024). These frameworks share a common aspiration: to ensure that AI systems are trustworthy, human-centred, and accountable. Yet a fundamental implementation gap persists. At the point where accountability matters most — the moment a clinical AI system influences a decision about a patient's care — none of these frameworks resolves the operational question of who, specifically, is responsible. This paper argues that global AI governance frameworks fail at the point of clinical decision-making for three structural reasons: they are architected around principles rather than named accountability; they were designed primarily for high-income country regulatory environments; and they address AI at a systemic level without specifying the clinical and institutional roles that must be accountable at the point of care. Drawing on the Clinical AI Accountability Framework developed in Working Paper No. 1 of this series (Adebiyi, 2026a), the paper argues that closing the implementation gap requires moving from principle-level governance to role-level accountability, and that African health systems are uniquely positioned to pioneer this shift.