Artificial intelligence (AI) is increasingly embedded in development systems, enabling new capabilities for poverty prediction, social protection targeting, and service delivery optimisation across sectors such as finance, agriculture, health and education, yet its implications for poverty governance in low- and middle-income settings remain fragmented. This study conducted a systematic literature review of South Africa’s DHET peer-reviewed journal articles and scholarly book chapters published within the last decade, screening studies for relevance to AI-enabled poverty reduction applications including predictive analytics, high-resolution poverty mapping, digital financial inclusion, precision agriculture, health diagnostics, educational personalisation, and public-sector digital transformation. A thematic synthesis was applied to identify cross-cutting patterns related to system performance, implementation processes, governance considerations, and contextual constraints. The reviewed evidence indicates that AI can improve poverty governance through multimodal data integration, enhanced targeting accuracy, automated administrative processes, expanded access to financial and basic services, and strengthened rural livelihood systems. However, persistent challenges include biased or incomplete datasets, infrastructural and computational limitations, weak interoperability, regulatory gaps, and ethical risks regarding privacy, accountability and exclusion, which may reinforce structural inequalities through misclassification and unequal access. The review contributes an integrated evidence base and highlights that developmental gains from AI depend on robust data governance, inclusive digital infrastructure, context-sensitive design, algorithmic transparency, and institutional capacity, while future research should prioritise impact evaluation, fairness-aware and explainable AI, participatory design, and scalable approaches for low-resource environments.