<p>Effective disaster risk reduction depends not only on hazard detection, but also on the timely dissemination of administrative early warnings. However, datasets that systematically document how warnings are issued across administrative hierarchies remain limited. Here, we present the Chinese Multi-Hazard Early Warning Dataset, a standardized database containing 1,057,817 warning records issued across mainland China between 1 January 2022 and 31 December 2025. The dataset integrates records from the national emergency broadcasting platform and official social media accounts operated within China’s warning dissemination system. Raw warning texts were processed through a standardized workflow including spatiotemporal geocoding, hierarchical hazard classification, and record cleaning and integration. The final dataset provides hourly records across four administrative tiers (national, provincial, city-level, and county-level) using a unified taxonomy of hydrometeorological, geophysical, oceanographic, and environmental hazards, each graded according to four levels of severity. Comparisons with official statistical reports show strong temporal agreement across administrative levels (r &gt; 0.94, p &lt; 0.001), supporting the overall credibility of the dataset. This dataset complements traditional hazard observations by documenting the operational layer of warning dissemination and provides empirical resources for studying warning governance, dissemination timing, administrative coordination, and risk communication.</p>