Bridging the Multilingual Safety Divide: Efficient, Culturally-Aware Alignment for Global South Languages

2026-02-14

AI Summary

Current safety mechanisms in large language models significantly underperform in low-resource languages and code-mixed settings prevalent in the Global South, and often fail to align with local cultural understandings of “harmful content.” This work proposes the first efficient, culturally sensitive safety alignment framework tailored for the Global South, integrating parameter-efficient fine-tuning, culturally contextualized preference data collection, a multilingual safety evaluation benchmark, and a community-informed workflow for defining harms. Moving beyond English-centric paradigms, the framework shifts multilingual AI safety from mere technical adaptation toward collaborative cultural co-construction, offering a systematic research pathway toward equitable and actionable localized safety systems.

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
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Large language models (LLMs) are being deployed across the Global South, where everyday use involves low-resource languages, code-mixing, and culturally specific norms. Yet safety pipelines, benchmarks, and alignment still largely target English and a handful of high-resource languages, implicitly assuming safety and factuality''transfer''across languages. Evidence increasingly shows they do not. We synthesize recent findings indicating that (i) safety guardrails weaken sharply on low-resource and code-mixed inputs, (ii) culturally harmful behavior can persist even when standard toxicity scores look acceptable, and (iii) English-only knowledge edits and safety patches often fail to carry over to low-resource languages. In response, we outline a practical agenda for researchers and students in the Global South: parameter-efficient safety steering, culturally grounded evaluation and preference data, and participatory workflows that empower local communities to define and mitigate harm. Our aim is to make multilingual safety a core requirement-not an add-on-for equitable AI in underrepresented regions.

Problem

• multilingual safety
• low-resource languages
• code-mixing
• cultural harm
• alignment

Innovation

• parameter-efficient safety steering
• culturally grounded evaluation
• code-mixed inputs
• multilingual alignment
• participatory AI