UN turns to Google to make its global data ready for AI agents

UNICEF’s benchmark test revealed a critical flaw in leading AI models. Six large language models, including GPT-4o and Gemini 2.5 Flash, achieved only a 21.2% average accuracy in retrieving global development statistics. About three in five responses provided no usable number. This reflects a persistent hallucination problem, even with authoritative data sources.

The UN’s partnership with Google, backed by $2 million from Google.org, aims to address this data reliability issue. The UN System Data Commons will host data from nearly 20 UN entities at launch. This move will benefit Asian AI developers and researchers who rely on accurate global data. It also sets a standard for other international organizations to make their data AI-ready.

The thing to watch is the UN’s commitment to bring 80% of its statistical datasets onto the platform by 2027. Success here would validate Google’s Data Commons as a critical infrastructure layer for reliable AI. Failure would show that even with dedicated efforts, AI still struggles with authoritative data at scale.