Media Framing of AI Across Global Powers

Public conversation about AI has accelerated since 2024, yet we lack cross-lingual, comparative evidence of how major news outlets across global powers frame AI. In this work, we study cross-national media framing of AI by constructing a multilingual dataset of 1,970 news articles published between January 2024 and October 2025 from 47 outlets across the U.S., China, and Russia. We introduce a taxonomy of 25 AI-specific frames with sentiment labels and perform sentence-level human annotation to capture how AI is discussed across languages. An analysis of our dataset reveals substantial cross-country differences: Chinese media consistently portray AI more positively, while U.S. and Russian coverage is more mixed; work-related frames dominate coverage, whereas safety-related concerns appear less frequently. We also observe frequent cross-national discourse, with media outlets often referencing other countries in competitive contexts. Using our data, we evaluate large language models (LLMs) on the framing classification task and show gaps in LLM performance, particularly in frame identification, highlighting the need for human-centered analysis of multilingual media narratives.