IndicSafe: A Benchmark for Evaluating Multilingual LLM Safety in South Asia

TL;DR

This paper introduces IndicSafe, a benchmark for evaluating the safety of multilingual large language models across 12 Indic languages, revealing significant safety inconsistencies and gaps in current models.

Contribution

It presents the first systematic safety evaluation across Indic languages, highlighting safety drift and proposing a benchmark for culturally informed safety assessment.

Findings

Cross-language safety agreement is only 12.8%.

Safety rate variance exceeds 17% across languages.

Models show over-refusal and over-flagging issues.

Abstract

As large language models (LLMs) are deployed in multilingual settings, their safety behavior in culturally diverse, low-resource languages remains poorly understood. We present the first systematic evaluation of LLM safety across 12 Indic languages, spoken by over 1.2 billion people but underrepresented in LLM training data. Using a dataset of 6,000 culturally grounded prompts spanning caste, religion, gender, health, and politics, we assess 10 leading LLMs on translated variants of the prompt. Our analysis reveals significant safety drift: cross-language agreement is just 12.8\%, and \texttt{SAFE} rate variance exceeds 17\% across languages. Some models over-refuse benign prompts in low-resource scripts, overflag politically sensitive topics, while others fail to flag unsafe generations. We quantify these failures using prompt-level entropy, category bias scores, and multilingual consistency indices. Our findings highlight critical safety generalization gaps in multilingual LLMs and show that safety alignment does not transfer evenly across languages. We release \textsc{IndicSafe}, the first benchmark to enable culturally informed safety evaluation for Indic deployments, and advocate for language-aware alignment strategies grounded in regional harms.