Geopolitical Bias in Sovereign Large Language Models: A Comparative Mixed-Methods Study

Sovereign large language models (LLMs), emerging as strategic assets in global information ecosystems, represent advanced AI system developed under distinct national governance regimes. This study examines how model origin and governance context influence AI-generated narratives on international territorial disputes. The study compares outputs from three prominent sovereign LLMs - OpenAI’s GPT-4o (United States), DeepSeek-R1 (China), and Mistral (European Union), across 12 high-profile territorial conflicts. Statistically significant differences in each model's sentiment distribution and geopolitical framing are identified using a mixed-methods approach that combines sentiment analysis with statistical evaluation (chi-square tests and analysis of variance, ANOVA) on responses to 300 standardized prompts. The findings indicate model provenance substantially shapes the tone and stance of outputs, with each LLM reflecting distinct biases aligned with its national context. These disparities carry important policy and societal implications: reliance on a single sovereign model could inadvertently bias public discourse and decision-making toward that model's native perspective. The study highlights ethical considerations such as transparency and fairness and calls for robust governance frameworks. It underscores the need for careful oversight and international cooperation to ensure that sovereign LLMs are deployed in a manner that supports informed and balanced geopolitical dialogue.© The Author(s) 2025. Published by RITHA Publishing. This article is distributed under the terms of the license CC-BY 4.0., which permits any further distribution in any medium, provided the original work is properly cited maintaining attribution to the author(s) and the title of the work, journal citation and URL DOI.