English

Soteria: Language-Specific Functional Parameter Steering for Multilingual Safety Alignment

Computation and Language 2025-08-25 v2 Artificial Intelligence

Abstract

Ensuring consistent safety across multiple languages remains a significant challenge for large language models (LLMs). We introduce Soteria, a lightweight yet powerful strategy that locates and minimally adjusts the "functional heads" most responsible for harmful content generation in each language. By altering only a fraction of parameters, Soteria drastically reduces policy violations without sacrificing overall model performance, even in low-resource settings. To rigorously evaluate our approach, we also present XThreatBench, a specialized multilingual dataset capturing fine-grained harmful behaviors drawn from real policy guidelines. Experiments with leading open-source LLMs (e.g., Llama, Qwen, Mistral) show that Soteria consistently improves safety metrics across high-, mid-, and low-resource languages. These findings highlight a promising path toward scalable, linguistically attuned, and ethically aligned LLMs worldwide.

Keywords

Cite

@article{arxiv.2502.11244,
  title  = {Soteria: Language-Specific Functional Parameter Steering for Multilingual Safety Alignment},
  author = {Somnath Banerjee and Sayan Layek and Pratyush Chatterjee and Animesh Mukherjee and Rima Hazra},
  journal= {arXiv preprint arXiv:2502.11244},
  year   = {2025}
}

Comments

Accepted at EMNLP 2025

R2 v1 2026-06-28T21:46:12.372Z