IndicGuard: A Multilingual Safety Guard Model and Dataset for Indic Languages

2026-06-22

AI Summary

This work addresses the critical gap in safety mechanisms for large language models, which are predominantly optimized for English and often fail to recognize culturally sensitive or localized harmful content in Indic languages. The study presents the first systematic effort to construct a high-quality, culturally aware safety dataset spanning ten major Indic languages. Building upon Gemma-3-4B-IT, the authors train a 4-billion-parameter multilingual safety guardrail model through instruction tuning, adversarial example augmentation, and cross-lingual transfer learning. The resulting model significantly outperforms the current state-of-the-art baseline, CultureGuard, across all evaluated languages, demonstrating strong cross-lingual generalization capabilities. It effectively detects localized adversarial jailbreak attacks and successfully extends its safety coverage to unseen, low-resource Indic languages.

Abstract

As Large Language Models (LLMs) achieve widespread integration across diverse linguistic landscapes, ensuring their safety and alignment with regional normative values remains a critical challenge. Current safety mechanisms are predominantly optimized for English-centric frameworks, often failing to capture the unique socio-cultural sensitivities and localized categories of harm inherent to the Indic region. To address this gap, we introduce IndicGuard, a multilingual safety guard model and dataset for Indic languages. We construct a high-volume, culturally nuanced safety dataset encompassing ten major Indic languages, systematically curated to capture regional harms, sensitive socio-political contexts, and adversarial jailbreaks. Leveraging this corpus, we fine-tune a 4B-parameter instruction-tuned model based on Gemma-3-4B-IT to serve as a multilingual safety guardrail for real-time content moderation and policy compliance checking. Our empirical evaluations demonstrate that IndicGuard significantly enhances LLM robustness against localized vulnerabilities, achieving high moderation consistency across different conversational turns. Crucially, IndicGuard consistently outperforms the existing baseline model, CultureGuard, across evaluated languages. Finally, we demonstrate that our model effectively generalizes to low-resource Indic languages excluded from training, substantiating the structural robustness and cross-lingual transfer capabilities of the framework.

Problem

• safety alignment
• Indic languages
• multilingual LLMs
• cultural sensitivity
• localized harm

Innovation

• multilingual safety guard
• Indic languages
• culturally nuanced dataset
• cross-lingual transfer
• adversarial jailbreak detection