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

2026-01-30

AI Summary

This study addresses the lack of systematic evaluation and cross-lingual consistency control for political ideological bias in large language models, particularly in low-resource and non-Western contexts where safe and effective post-hoc mitigation mechanisms are scarce. To this end, the authors construct a large-scale benchmark covering 33 languages across 50 countries and propose the Cross-Lingual Alignment Steering (CLAS) framework. CLAS aligns implicit ideological representations from diverse languages into a shared subspace, enabling adaptive intervention strength for consistent bias mitigation across languages. Experimental results demonstrate that the method significantly reduces political bias along both economic and social dimensions while preserving generation quality almost entirely. This work thus offers a scalable and interpretable paradigm for fair governance of multilingual large language models.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Bias, Fairness & Equity

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

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.

Problem

• political bias
• multilingual LLMs
• cross-lingual consistency
• ideological neutrality
• fairness

Innovation

• Cross-Lingual Alignment Steering
• multilingual political bias
• ideological representation alignment
• adaptive bias mitigation
• fairness-aware LLM governance