The Multilingual Divide and Its Impact on Global AI Safety

2025-05-27

AI Summary

This paper identifies a critical multilingual capability gap in global AI safety: mainstream large language models exhibit systematic performance degradation and heightened security vulnerabilities in non-dominant languages—particularly African, Southeast Asian, and Indigenous languages. Methodologically, the study establishes the first causal framework linking “linguistic disparity” to “security inequity,” integrating cross-lingual safety evaluation, multi-source modeling of linguistic capability gaps, and policy pathway simulation. It identifies three structural barriers: data scarcity, absence of standardized multilingual safety benchmarks, and insufficient model transparency. The contribution is a pragmatic, governance-oriented framework for multilingual AI safety, comprising (1) development of high-quality, diverse multilingual datasets; (2) open-sourced, cross-lingual safety evaluation benchmarks; and (3) a multilateral collaborative governance mechanism. The findings provide both theoretical grounding and actionable policy levers to address global disparities in AI safety capacity.

Technology Category

• Natural Language Processing: Safety and Robustness
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Security and Privacy: Security and privacy of machine learning and AI applications
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Despite advances in large language model capabilities in recent years, a large gap remains in their capabilities and safety performance for many languages beyond a relatively small handful of globally dominant languages. This paper provides researchers, policymakers and governance experts with an overview of key challenges to bridging the"language gap"in AI and minimizing safety risks across languages. We provide an analysis of why the language gap in AI exists and grows, and how it creates disparities in global AI safety. We identify barriers to address these challenges, and recommend how those working in policy and governance can help address safety concerns associated with the language gap by supporting multilingual dataset creation, transparency, and research.

Problem

• Addressing the AI capability gap for non-dominant languages
• Reducing safety risks disparities across different languages
• Promoting multilingual datasets and research transparency

Innovation

• Support multilingual dataset creation
• Enhance transparency in AI models
• Promote research on language disparities