Researchers have released a large, open dataset of state media articles, covering 118 million pieces from 136 countries. The collection is curated, annotated, and multilingual, offering a structured view of how governments project narratives to domestic and foreign audiences.
The dataset addresses a growing concern: state media output is plentiful and freely accessible, so it shows up widely in the web corpora that large language models are trained on. By making this content easier to study, the resource could help researchers assess the influence of state-backed messaging on AI systems.
The abstract notes that state media often targets both domestic and foreign audiences, and the dataset appears designed to support analysis of those dynamics. No further details on methodology or annotation schemes are provided in the available summary.