This groundbreaking research systematically examines geopolitical bias across major LLMs using innovative bilingual and dual-framing methodology to evaluate responses on U.S.-China relations.

Key findings:

• Evaluated 11 prominent LLMs using 19,712 prompts in both English and Chinese
• Quantitatively assessed bias on a scale from strongly Pro-China (-2) to strongly Pro-U.S. (+2)
• Revealed how language models respond differently to politically sensitive topics
• Demonstrated how biases can affect AI systems in security and international contexts

Security Implications: Understanding these biases is crucial for national security applications where AI systems interpret or generate content related to geopolitical tensions. Organizations must recognize these inherent leanings when deploying LLMs in sensitive contexts.

Mapping Geopolitical Bias in 11 Large Language Models: A Bilingual, Dual-Framing Analysis of U.S.-China Tensions