Who Transfers Safety? Identifying and Targeting Cross-Lingual Shared Safety Neurons

2026-02-01

AI Summary

This work addresses the pronounced imbalance in safety alignment across languages in multilingual large language models, where low-resource languages exhibit substantially weaker safety guarantees compared to high-resource ones, and the underlying neural mechanisms remain unclear. The study identifies, for the first time, a class of cross-lingually shared safety neurons (SS-Neurons) and validates their pivotal role in transferring safety capabilities through causal intervention techniques. Building on this insight, the authors propose a targeted training strategy that fine-tunes only this minimal subset of neurons. This approach significantly enhances safety in low-resource languages while preserving the model’s general capabilities, outperforming current state-of-the-art multilingual safety alignment methods.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Security and Privacy: Large-scale security measurements

Abstract

Multilingual safety remains significantly imbalanced, leaving non-high-resource (NHR) languages vulnerable compared to robust high-resource (HR) ones. Moreover, the neural mechanisms driving safety alignment remain unclear despite observed cross-lingual representation transfer. In this paper, we find that LLMs contain a set of cross-lingual shared safety neurons (SS-Neurons), a remarkably small yet critical neuronal subset that jointly regulates safety behavior across languages. We first identify monolingual safety neurons (MS-Neurons) and validate their causal role in safety refusal behavior through targeted activation and suppression. Our cross-lingual analyses then identify SS-Neurons as the subset of MS-Neurons shared between HR and NHR languages, serving as a bridge to transfer safety capabilities from HR to NHR domains. We observe that suppressing these neurons causes concurrent safety drops across NHR languages, whereas reinforcing them improves cross-lingual defensive consistency. Building on these insights, we propose a simple neuron-oriented training strategy that targets SS-Neurons based on language resource distribution and model architecture. Experiments demonstrate that fine-tuning this tiny neuronal subset outperforms state-of-the-art methods, significantly enhancing NHR safety while maintaining the model's general capabilities. The code and dataset will be available athttps://github.com/1518630367/SS-Neuron-Expansion.

Problem

• multilingual safety
• cross-lingual transfer
• safety alignment
• low-resource languages
• neural mechanisms

Innovation

• cross-lingual safety transfer
• shared safety neurons
• neuron-oriented training
• multilingual alignment
• safety fine-tuning