When Safety Speaks a Language: A Mechanistic Analysis of Safety-Language Identity Entanglement in LLMs

• URL: http://arxiv.org/abs/2608.29936v1
• Date: Sun, 30 Aug 2026 18:07:26 GMT
• Title: When Safety Speaks a Language: A Mechanistic Analysis of Safety-Language Identity Entanglement in LLMs
• Abstract summary: Safety-relevant features are architecture-dependent in terms of where they are located and how they are distributed across layers.<n>This safety-language entanglement has direct consequences such that ablating safety features impacts not only harmful response rates but also target language.<n>Our findings qualify the language-universality of safety alignment as architecture-dependent and offer a mechanistic account of multilingual safety interventions.
• Score: 4.92475688183711
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Safety alignment of large language models (LLMs) degrades across languages, yet the internal mechanism driving this asymmetry remains poorly understood. Our work, therefore, presents a systematic mechanistic analysis of multilingual safety using sparse autoencoder (SAE) features, sparse interpretable directions in the residual stream associated with harmful and harmless model behavior across three instruction-tuned LLMs, eight languages, and all model layers. We observe that safety-relevant features are architecture-dependent in terms of where they are located and how they are distributed across layers. Additionally, they are geometrically entangled with language identity and exhibit cross-lingual sharing patterns, i.e., languages share safety features to varying degrees across model depths and architectures. This safety-language entanglement has direct consequences such that ablating safety features impacts not only harmful response rates but also target language, with the degree of intervention predicted by the relationship between safety and language features. Our findings qualify the language-universality of safety alignment as architecture-dependent and offer a mechanistic account of multilingual safety interventions.

Related papers

• Who Bridges Safety? Identifying and Targeting Cross-Lingual Shared Safety Pathways [44.23839982612062]
  We identify and target the cross-layer functional pathways formed during safety signal propagation.<n>We propose a pathways-targeted alignment method based on the cross-lingual shared safety pathways.<n> Experimental results show that updating only a small fraction of pathway parameters significantly improves safety in NHR languages.
  arXiv  Detail & Related papers  (2026-08-10T03:48:47Z)
• Exploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language Models [37.24612885977265]
  Multimodal Large Language Models integrate visual perception into language reasoning, introducing a continuous attack surface susceptible to adversarial attacks.<n>Prior work on MLLM has focused largely on English-centric tasks, leaving multilingual behaviour unexplored.<n>We study adversarial robustness and multimodal safety across 12 diverse languages, evaluating open-source MLLMs that acquire multilingual capability through instruction tuning.
  arXiv  Detail & Related papers  (2026-06-02T15:42:10Z)
• Multilingual Safety Alignment via Self-Distillation [17.94152626632751]
  Large language models (LLMs) exhibit severe multilingual safety misalignment.<n>We propose a cross-lingual safeguard transfer framework named Multilingual Self-Distillation (MSD)<n>Our framework is flexible and can be integrated with different self-distillation strategies.
  arXiv  Detail & Related papers  (2026-05-03T14:22:49Z)
• Multilingual Refusal Alignment for Safer Large Language Models [53.64286756804503]
  We investigate whether single-language alignment transfers cross-lingually, how language consistency is preserved during training, and the resulting trade-offs with general knowledge capabilities.<n>We introduce RefusEU, a novel refusal alignment dataset covering 12 European languages, including a dedicated test set for evaluating current state-of-the-art models.<n>Our controlled Direct Preference Optimization (DPO) experiments provide two key insights: aligning models exclusively in English is insufficient to ensure cross-lingual safety, even for the same harm categories, whereas training on multilingual datasets can improve safety without degrading general performance, as measured by the Global MMLU benchmark.
  arXiv  Detail & Related papers  (2026-04-24T09:29:14Z)
• LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety [88.98698230989186]
  Large language models (LLMs) often demonstrate strong safety performance in high-resource languages, yet exhibit severe vulnerabilities in low-resource languages.<n>We attribute this gap to a mismatch between language-agnostic semantic understanding ability and language-dominant safety alignment biased toward high-resource languages.<n>We propose Language-Agnostic Semantic Alignment (LASA), which anchors safety alignment directly in semantic bottlenecks.
  arXiv  Detail & Related papers  (2026-04-13T15:59:50Z)
• Precise Shield: Explaining and Aligning VLLM Safety via Neuron-Level Guidance [64.18179414483966]
  In real-world deployments, Vision-Language Large Models (VLLMs) face critical challenges from multilingual and multimodal composite attacks.<n>We propose Precise Shield, a two-stage framework that first identifies safety neurons by contrasting activation patterns between harmful and benign inputs.<n>This strategy substantially improves safety while preserving multilingual and multimodal generalization.
  arXiv  Detail & Related papers  (2026-04-10T02:42:52Z)
• LinguaSafe: A Comprehensive Multilingual Safety Benchmark for Large Language Models [22.273388934888278]
  Our dataset comprises 45k entries in 12 languages, ranging from Hungarian to Malay.<n>Our benchmark provides a comprehensive suite of metrics for in-depth safety evaluation.
  arXiv  Detail & Related papers  (2025-08-18T08:59:01Z)
• MPO: Multilingual Safety Alignment via Reward Gap Optimization [88.76638442683391]
  Large language models (LLMs) have become increasingly central to AI applications worldwide.<n>Existing preference learning methods for safety alignment, such as RLHF and DPO, are primarily monolingual and struggle with noisy multilingual data.<n>We introduce Multilingual reward gaP Optimization (MPO), a novel approach that leverages the well-aligned safety capabilities of the dominant language (English) to improve safety alignment across multiple languages.
  arXiv  Detail & Related papers  (2025-05-22T16:24:51Z)
• MrGuard: A Multilingual Reasoning Guardrail for Universal LLM Safety [56.77103365251923]
  Large Language Models (LLMs) are susceptible to adversarial attacks such as jailbreaking.<n>This vulnerability is exacerbated in multilingual settings, where multilingual safety-aligned data is often limited.<n>We introduce a multilingual guardrail with reasoning for prompt classification.
  arXiv  Detail & Related papers  (2025-04-21T17:15:06Z)
• LLMs Lost in Translation: M-ALERT uncovers Cross-Linguistic Safety Inconsistencies [63.10843814055688]
  M-ALERT is a benchmark that evaluates the safety of Large Language Models in five languages.<n>M-ALERT includes 15k high-quality prompts per language, totaling 75k, with category-wise annotations.<n>Our experiments on 39 state-of-the-art LLMs highlight the importance of language-specific safety analysis.
  arXiv  Detail & Related papers  (2024-12-19T16:46:54Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.