Bias Beyond Borders: Political Ideology Evaluation and Steering in Multilingual LLMs

Large Language Models (LLMs) increasingly shape global discourse, making fairness and ideological neutrality essential for responsible AI deployment. Despite growing attention to political bias in LLMs, prior work largely focuses on high-resource, Western languages or narrow multilingual settings, leaving cross-lingual consistency and safe post-hoc mitigation underexplored. To address this gap, we present a large-scale multilingual evaluation of political bias spanning 50 countries and 33 languages. We introduce a complementary post-hoc mitigation framework, Cross-Lingual Alignment Steering (CLAS), designed to augment existing steering methods by aligning ideological representations across languages and dynamically regulating intervention strength. This method aligns latent ideological representations induced by political prompts into a shared ideological subspace, ensuring cross lingual consistency, with the adaptive mechanism prevents over correction and preserves coherence. Experiments demonstrate substantial bias reduction along both economic and social axes with minimal degradation in response quality. The proposed framework establishes a scalable and interpretable paradigm for fairness-aware multilingual LLM governance, balancing ideological neutrality with linguistic and cultural diversity.

💡 Research Summary

The paper addresses the pressing need for fairness and ideological neutrality in large language models (LLMs) that increasingly shape global discourse. While prior work on political bias has largely focused on high‑resource, Western languages, this study expands the scope to a truly multilingual setting, evaluating 50 countries across 33 languages using the Political Compass Test (PCT). The PCT decomposes political stance into two orthogonal axes—economic (left‑right) and social (libertarian‑authoritarian)—and the authors adapt it for multilingual use by translating and culturally adapting each statement. Responses are collected on a four‑point Likert scale, converted into a continuous stance score S∈