Refusal Direction is Universal Across Safety-Aligned Languages

• URL: http://arxiv.org/abs/2505.17306v1
• Date: Thu, 22 May 2025 21:54:46 GMT
• Title: Refusal Direction is Universal Across Safety-Aligned Languages
• Authors: Xinpeng Wang, Mingyang Wang, Yihong Liu, Hinrich Schütze, Barbara Plank,
• Abstract summary: In this paper, we investigate the refusal behavior in large language models (LLMs) across 14 languages using PolyRefuse.<n>We uncover the surprising cross-lingual universality of the refusal direction: a vector extracted from English can bypass refusals in other languages with near-perfect effectiveness.<n>We attribute this transferability to the parallelism of refusal vectors across languages in the embedding space and identify the underlying mechanism behind cross-lingual jailbreaks.
• Score: 66.64709923081745
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Refusal mechanisms in large language models (LLMs) are essential for ensuring safety. Recent research has revealed that refusal behavior can be mediated by a single direction in activation space, enabling targeted interventions to bypass refusals. While this is primarily demonstrated in an English-centric context, appropriate refusal behavior is important for any language, but poorly understood. In this paper, we investigate the refusal behavior in LLMs across 14 languages using PolyRefuse, a multilingual safety dataset created by translating malicious and benign English prompts into these languages. We uncover the surprising cross-lingual universality of the refusal direction: a vector extracted from English can bypass refusals in other languages with near-perfect effectiveness, without any additional fine-tuning. Even more remarkably, refusal directions derived from any safety-aligned language transfer seamlessly to others. We attribute this transferability to the parallelism of refusal vectors across languages in the embedding space and identify the underlying mechanism behind cross-lingual jailbreaks. These findings provide actionable insights for building more robust multilingual safety defenses and pave the way for a deeper mechanistic understanding of cross-lingual vulnerabilities in LLMs.

Related papers

• MrGuard: A Multilingual Reasoning Guardrail for Universal LLM Safety [56.79292318645454]
  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)
• Lost in Multilinguality: Dissecting Cross-lingual Factual Inconsistency in Transformer Language Models [49.16690802656554]
  We find that Multilingual factual models struggle to provide consistent responses to semantically equivalent prompts in different languages.<n>We propose a linear shortcut method that bypasses computations in the final layers, enhancing both prediction accuracy and cross-lingual consistency.
  arXiv  Detail & Related papers  (2025-04-05T19:43:10Z)
• The Hidden Space of Safety: Understanding Preference-Tuned LLMs in Multilingual context [0.9130277390156759]
  Alignment tuning has enabled large language models to excel in reasoning, instruction-following, and minimizing harmful generations.<n>Despite their widespread deployment, these models exhibit a monolingual bias, raising concerns about the effectiveness of alignment across languages.<n>Current alignment methods predominantly focus on English, leaving it unclear how alignment mechanism generalizes to multilingual settings.
  arXiv  Detail & Related papers  (2025-04-03T15:46:46Z)
• Benchmarking LLM Guardrails in Handling Multilingual Toxicity [57.296161186129545]
  We introduce a comprehensive multilingual test suite, spanning seven datasets and over ten languages, to benchmark the performance of state-of-the-art guardrails. We investigate the resilience of guardrails against recent jailbreaking techniques, and assess the impact of in-context safety policies and language resource availability on guardrails' performance. Our findings show that existing guardrails are still ineffective at handling multilingual toxicity and lack robustness against jailbreaking prompts.
  arXiv  Detail & Related papers  (2024-10-29T15:51:24Z)
• Towards Understanding the Fragility of Multilingual LLMs against Fine-Tuning Attacks [18.208272960774337]
  Large Language Models (LLMs) have sparked widespread concerns about their safety.<n>Recent work demonstrates that safety alignment of LLMs can be easily removed by fine-tuning.<n>We take a further step to understand fine-tuning attacks in multilingual LLMs.
  arXiv  Detail & Related papers  (2024-10-23T18:27:36Z)
• Against All Odds: Overcoming Typology, Script, and Language Confusion in Multilingual Embedding Inversion Attacks [3.2297018268473665]
  Large Language Models (LLMs) are susceptible to malicious influence by cyber attackers through intrusions such as adversarial, backdoor, and embedding inversion attacks.<n>This study explores the security of multilingual LLMs in the context of embedding inversion attacks and investigates cross-lingual and cross-script inversion across 20 languages.<n>Our findings indicate that languages written in Arabic script and Cyrillic script are particularly vulnerable to embedding inversion, as are languages within the Indo-Aryan language family.
  arXiv  Detail & Related papers  (2024-08-21T16:16:34Z)
• TuBA: Cross-Lingual Transferability of Backdoor Attacks in LLMs with Instruction Tuning [63.481446315733145]
  Cross-lingual backdoor attacks against multilingual large language models (LLMs) are under-explored.<n>Our research focuses on how poisoning the instruction-tuning data for one or two languages can affect the outputs for languages whose instruction-tuning data were not poisoned.<n>Our method exhibits remarkable efficacy in models like mT5 and GPT-4o, with high attack success rates, surpassing 90% in more than 7 out of 12 languages.
  arXiv  Detail & Related papers  (2024-04-30T14:43:57Z)
• Text Embedding Inversion Security for Multilingual Language Models [2.790855523145802]
  Research shows that text can be reconstructed from embeddings, even without knowledge of the underlying model. This study is the first to investigate multilingual inversion attacks, shedding light on the differences in attacks and defenses across monolingual and multilingual settings.
  arXiv  Detail & Related papers  (2024-01-22T18:34:42Z)
• Inducing Language-Agnostic Multilingual Representations [61.97381112847459]
  Cross-lingual representations have the potential to make NLP techniques available to the vast majority of languages in the world. We examine three approaches for this: (i) re-aligning the vector spaces of target languages to a pivot source language; (ii) removing language-specific means and variances, which yields better discriminativeness of embeddings as a by-product; and (iii) increasing input similarity across languages by removing morphological contractions and sentence reordering.
  arXiv  Detail & Related papers  (2020-08-20T17:58:56Z)

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.