mlsanityData to 13 Oct 2026

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

arXiv 2608.09095 ↗First posted 10 Aug 2026cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

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.

Key facts. No citing papers in the map yet (data through 13 Oct 2026). Builds on 27 of its 38 references that are in the map. First posted on arXiv 10 Aug 2026.

Builds on

27 of 38 references.

References · 27

  1. 2025Qwen3 Technical Reportcited by 8,510
  2. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,986
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,876
  4. 2022Locating and Editing Factual Associations in GPTcited by 2,650
  5. 2020Transformer Feed-Forward Layers Are Key-Value Memoriescited by 1,323
  6. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,141
  7. 2022Mass-Editing Memory in a Transformercited by 990
  8. 2024SimPO: Simple Preference Optimization with a Reference-Free Rewardcited by 954
  9. 2024ORPO: Monolithic Preference Optimization without Reference Modelcited by 558
  10. 2022Language Models are Multilingual Chain-of-Thought Reasonerscited by 525
  11. 2023Low-Resource Languages Jailbreak GPT-4cited by 306
  12. 2023Efficient and Effective Text Encoding for Chinese LLaMA and Alpacacited by 304
15 more
  1. 2024Do Llamas Work in English? On the Latent Language of Multilingual Transformerscited by 280
  2. 2024KTO: Model Alignment as Prospect Theoretic Optimizationcited by 259
  3. 2024Provably Robust DPO: Aligning Language Models with Noisy Feedbackcited by 134
  4. 2025Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrailscited by 128
  5. 2024The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contextscited by 92
  6. 2023Cognitive Overload: Jailbreaking Large Language Models with Overloaded Logical Thinkingcited by 81
  7. 2024Enhancing Multilingual Capabilities of Large Language Models through Self-Distillation from Resource-Rich Languagescited by 36
  8. 2025CultureGuard: Towards Culturally-Aware Dataset and Guard Model for Multilingual Safety Applicationscited by 25
  9. 2025Beyond Prompt Engineering: Robust Behavior Control in LLMs via Steering Target Atomscited by 23
  10. 2023InstructAlign: High-and-Low Resource Language Alignment via Continual Crosslingual Instruction Tuningcited by 21
  11. 2025The Emergence of Abstract Thought in Large Language Models Beyond Any Languagecited by 20
  12. 2025LinguaSafe: A Comprehensive Multilingual Safety Benchmark for Large Language Modelscited by 14
  13. 2024Popularity-Aware Alignment and Contrast for Mitigating Popularity Biascited by 9
  14. 2026Who Transfers Safety? Identifying and Targeting Cross-Lingual Shared Safety Neuronscited by 7
  15. 2024Know Your Account: Double Graph Inference-based Account De-anonymization on Ethereumcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons
  2. 2026When Safety Speaks a Language: A Mechanistic Analysis of Safety-Language Identity Entanglement in LLMs
  3. 2026SafeEvo: Deciphering the Safety Alignment Mechanism and Evolution in Language Models
  4. 2026SafeNexus: Discovering and Steering Modality-Universal Safety Neurons in MLLMs
  5. 2026One Anchor for All: Unified Multilingual and Multimodal Safety Alignment for LVLMs
  6. 2026BabelSteering: Multilingual Safety Alignment via English Steering Vectors
  7. 2026Multilingual Safety Signals Are Multi-Layered: Filtering Safety-Degrading Data for Safer LLMs
  8. 2026ACTR: Aligning Thoughts and Responses for Multilingual Safety in Reasoning LLMs