Geopolitical biases in LLMs: what are the "good" and the "bad" countries according to contemporary language models

This paper evaluates geopolitical biases in LLMs with respect to variouscountries though an analysis of their interpretation of historical events withconflicting national perspectives (USA, UK, USSR, and China). We introduce anovel dataset with neutral event descriptions and contrasting viewpoints fromdifferent countries. Our findings show significant geopolitical biases, withmodels favoring specific national narratives. Additionally, simple debiasingprompts had a limited effect in reducing these biases. Experiments withmanipulated participant labels reveal models' sensitivity to attribution,sometimes amplifying biases or recognizing inconsistencies, especially withswapped labels. This work highlights national narrative biases in LLMs,challenges the effectiveness of simple debiasing methods, and offers aframework and dataset for future geopolitical bias research.