Who Bridges Safety? Identifying and Targeting Cross-Lingual Shared Safety Pathways

2026-08-09

AI Summary

This work addresses the inadequate safety performance of multilingual large language models on low-resource languages, which stems from a limited understanding of cross-lingual safety signal propagation mechanisms. Moving beyond conventional isolated neuron analyses, the study identifies— for the first time—shared safety-functional pathways within the model that span layers and languages, uncovering an intrinsic bridge through which safety capabilities transfer from high-resource to low-resource languages. Building on this insight, the authors propose a targeted alignment method that fine-tunes only a minimal subset of pathway parameters. This approach substantially enhances safety in low-resource languages while preserving the model’s general capabilities, establishing a new paradigm for parameter-efficient safety deployment.

Abstract

Uncovering the internal mechanisms underlying the safety capabilities of large language models (LLMs) is crucial for developing trustworthy artificial intelligence. Currently, mechanistic interpretability studies on multilingual safety are largely confined to local components, such as isolated neurons. However, this static and fragmented perspective overlooks the synergy among components and fails to elucidate how safety signals dynamically propagate within the model to drive safety decisions ultimately. In this work, we move beyond isolated neurons to identify and target the cross-layer functional pathways formed during safety signal propagation, thereby uncovering the mechanisms driving the cross-lingual safety gap. Specifically, we first identify monolingual safety pathways and validate their impact on refusing harmful requests. Subsequent cross-lingual analyses reveal a sparse subset of cross-lingual shared safety pathways, confirming that this intersection acts as the internal bridge transferring safety capabilities from high-resource (HR) languages to non-high-resource (NHR) languages. Building on these mechanistic findings, we propose a pathways-targeted alignment method based on the cross-lingual shared safety pathways. Experimental results show that updating only a small fraction of pathway parameters significantly improves safety in NHR languages while largely preserving the model's general capabilities.

Problem

• mechanistic interpretability
• multilingual safety
• cross-lingual safety gap
• safety pathways
• large language models

Innovation

• cross-lingual safety pathways
• mechanistic interpretability
• safety signal propagation
• pathways-targeted alignment
• multilingual LLM safety