ThaiSafetyBench: Assessing Language Model Safety in Thai Cultural Contexts
This work is significant for researchers and developers working on LLM safety, as it highlights critical vulnerabilities in current safety alignment methods for non-English, culturally specific contexts, specifically for the Thai language.
This paper addresses the lack of safety evaluation for large language models (LLMs) in non-English languages by introducing ThaiSafetyBench, a benchmark of 1,954 malicious prompts in Thai. Their evaluation of 24 LLMs revealed that closed-source models generally perform better, and Thai-specific cultural attacks have a higher Attack Success Rate (ASR) than general Thai attacks. They also developed ThaiSafetyClassifier, a DeBERTa-based model, achieving an 84.4% weighted F1 score for harmful response classification.
The safety evaluation of large language models (LLMs) remains largely centered on English, leaving non-English languages and culturally grounded risks underexplored. In this work, we investigate LLM safety in the context of the Thai language and culture and introduce ThaiSafetyBench, an open-source benchmark comprising 1,954 malicious prompts written in Thai. The dataset covers both general harmful prompts and attacks that are explicitly grounded in Thai cultural, social, and contextual nuances. Using ThaiSafetyBench, we evaluate 24 LLMs, with GPT-4.1 and Gemini-2.5-Pro serving as LLM-as-a-judge evaluators. Our results show that closed-source models generally demonstrate stronger safety performance than open-source counterparts, raising important concerns regarding the robustness of openly available models. Moreover, we observe a consistently higher Attack Success Rate (ASR) for Thai-specific, culturally contextualized attacks compared to general Thai-language attacks, highlighting a critical vulnerability in current safety alignment methods. To improve reproducibility and cost efficiency, we further fine-tune a DeBERTa-based harmful response classifier, which we name ThaiSafetyClassifier. The model achieves a weighted F1 score of 84.4%, matching GPT-4.1 judgments. We publicly release the fine-tuning weights and training scripts to support reproducibility. Finally, we introduce the ThaiSafetyBench leaderboard to provide continuously updated safety evaluations and encourage community participation. - ThaiSafetyBench HuggingFace Dataset: https://huggingface.co/datasets/typhoon-ai/ThaiSafetyBench - ThaiSafetyBench Github: https://github.com/trapoom555/ThaiSafetyBench - ThaiSafetyClassifier HuggingFace Model: https://huggingface.co/typhoon-ai/ThaiSafetyClassifier - ThaiSafetyBench Leaderboard: https://huggingface.co/spaces/typhoon-ai/ThaiSafetyBench-Leaderboard