Study: 73% of Agentic AI for Social Good Papers Omit Geographic Context

Researchers from the paper arXiv:2605.22995 analyzed 112 papers on agentic AI for social good published between 2015 and 2026. They uncovered a stark 'moral-geographic asymmetry': 73% of papers (82 of 112) specify no geographic context at all. The problem is worse for institutional and social-policy domains—papers aligned with SDG 16 (peace, justice, strong institutions) are the most common in the corpus yet have the lowest geographic-specification rate at just 13%. In contrast, health or physical/ecological SDG papers specify geography 37–40% of the time. This pattern suggests that developers treat institutional good as universal, ignoring local political, legal, and cultural realities.

Even more concerning, only 28 of 112 papers (25%) report any real-world deployment or small-scale test. The study identifies five accountability gaps—from missing stakeholder engagement to absent failure modes documentation—and proposes a minimal reporting standard to force more context-specific, participatory, and accountable design. The authors argue that without grounding in actual communities, agentic AI risks imposing a one-size-fits-all solution that may do more harm than good, especially in fragile or underserved regions.