Mark RadarMARK RADAR
About
EN
Sign in

UN, Google Launch AI-Ready Global Data Platform

2 reports · First detected 2026-09-18 · Last active 2026-09-18

United Nations agencies compile some of the world’s most authoritative statistics on health, poverty, education and climate, but incompatible formats and siloed systems can make cross-agency analysis take months. The access problem has become more urgent as users increasingly rely on AI for factual answers. In a UNICEF benchmark covering more than 133,000 responses from six large language models, average accuracy on global development indicators was only 21.2%, underscoring the need to ground AI outputs in traceable official data.

The UN and Google launched the UN System Data Commons on September 17, 2026, replacing the older UNData portal with an open-source, AI-ready knowledge graph and natural-language search. Data from nearly 20 UN entities was available at launch, while 26 had committed to participate. The platform supports Model Context Protocol, allowing AI agents to retrieve statistics and source links directly. Google.org provided $2 million for infrastructure and capacity building, and the UN aims to include 80% of system-wide statistical datasets by 2027.

All Coverage

2 original reports

The Backstory

The history behind this event
Google Launches Groundsource Framework to Turn Global News Into Disaster Data With Gemini AIfirst seen 2026-03-12 · 1 reports · similarity 0.69 · same topic: Google

Flash floods develop quickly and affect small areas, while conventional monitoring data often suffer from limited geographic coverage and sparse records. That leaves forecasting models without verifiable historical benchmarks. Google Research created Groundsource to use Gemini to extract and cross-check locations and dates from news reports worldwide, supporting hydrological research, urban planning, government warning systems and emergency response.

Google Research released Groundsource on March 12, 2026, after analyzing more than 5 million news articles. The resulting open-access dataset contains 2.6 million records of historical flash floods across more than 150 countries from 2000 to the present. Human review found that both the location and time were accurate in 60% of records, while 82% were already useful for practical analysis. Related models can provide urban flash-flood forecasts up to 24 hours in advance through Flood Hub.

Mark Radar|MARK RADAR

If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →

All times are in Taipei time (GMT+8)