Governing Agentic AI in Africa: A Systematic Literature Review of Ethical Challenges, Framework Gaps, and Ubuntu-Informed Governance Solutions

Abstract

Agentic AI systems pursue goals through multi-step planning, tool use, memory, and adaptive action, creating governance challenges that differ from those associated with conventional predictive or generative AI. This study systematically reviews governance evidence relevant to agentic AI in Africa and develops an Ubuntu-informed conceptual framework. Following PRISMA 2020, 91 records published from 2019 to April 2026 were analysed through quantitative content analysis and abductive thematic synthesis. The corpus comprises 12 agentic-AI-specific records, 3 agentic-adjacent records, and 76 broader AI-governance records used as contextual evidence. Five literature-derived gaps were identified: accountability across autonomous action chains, infrastructure-aware oversight, bias and representational harm, policy fragmentation, and community exclusion. These findings inform five proposed governance elements: communal impact assessment, collective consent and participation, infrastructure-contingent oversight, African dataset and bias auditing, and communal accountability. The framework is a literature-derived conceptual proposition, not an empirically validated governance instrument, and requires future case-study, expert-panel, and comparative policy testing.