mlsanityData to 7 Oct 2026

The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts

arXiv 2401.13136 ↗First posted 23 Jan 2024cs.CL cs.AIcited by 92

On the map: LLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks

Abstract (arXiv)

As the influence of large language models (LLMs) spans across global communities, their safety challenges in multilingual settings become paramount for alignment research. This paper examines the variations in safety challenges faced by LLMs across different languages and discusses approaches to alleviating such concerns. By comparing how state-of-the-art LLMs respond to the same set of malicious prompts written in higher- vs. lower-resource languages, we observe that (1) LLMs tend to generate unsafe responses much more often when a malicious prompt is written in a lower-resource language, and (2) LLMs tend to generate more irrelevant responses to malicious prompts in lower-resource languages. To understand where the discrepancy can be attributed, we study the effect of instruction tuning with reinforcement learning from human feedback (RLHF) or supervised finetuning (SFT) on the HH-RLHF dataset. Surprisingly, while training with high-resource languages improves model alignment, training in lower-resource languages yields minimal improvement. This suggests that the bottleneck of cross-lingual alignment is rooted in the pretraining stage. Our findings highlight the challenges in cross-lingual LLM safety, and we hope they inform future research in this direction.

Key facts. Cited by 92 papers in the map (data through 7 Oct 2026). Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 33 of its 41 references that are in the map. First posted on arXiv 23 Jan 2024.

Builds on

33 of 41 references.

Closest ancestors · 1

Same direction, newest first

  1. 2023Universal and Transferable Adversarial Attacks on Aligned Language Models27 Jul

Other references · 32

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2017Proximal Policy Optimization Algorithmscited by 14,441
  3. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
  4. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  5. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  6. 2022PaLM: Scaling Language Modeling with Pathwayscited by 5,304
  7. 2019Unsupervised Cross-lingual Representation Learning at Scalecited by 3,907
  8. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  9. 2022Constitutional AI: Harmlessness from AI Feedbackcited by 2,950
  10. 2022BLOOM: A 176B-Parameter Open-Access Multilingual Language Modelcited by 1,968
  11. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,565
  12. 2022No Language Left Behind: Scaling Human-Centered Machine Translationcited by 1,290
20 more
  1. 2022Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Taskscited by 858
  2. 2020UnifiedQA: Crossing Format Boundaries With a Single QA Systemcited by 475
  3. 2024Contrastive Preference Optimization: Pushing the Boundaries of LLM Performance in Machine Translationcited by 390
  4. 2023Multi-step Jailbreaking Privacy Attacks on ChatGPTcited by 320
  5. 2022Quark: Controllable Text Generation with Reinforced Unlearningcited by 236
  6. 2020Are All Languages Created Equal in Multilingual BERT?cited by 206
  7. 2019Cross-Lingual Ability of Multilingual BERT: An Empirical Studycited by 182
  8. 2023Do-Not-Answer: A Dataset for Evaluating Safeguards in LLMscited by 176
  9. 2023The Capacity for Moral Self-Correction in Large Language Modelscited by 171
  10. 2022On the Origin of Hallucinations in Conversational Models: Is it the Datasets or the Models?cited by 137
  11. 2021AMMUS : A Survey of Transformer-based Pretrained Models in Natural Language Processingcited by 131
  12. 2021Systematic Inequalities in Language Technology Performance across the World's Languagescited by 118
  13. 2023Aligning Language Models with Preferences through f-divergence Minimizationcited by 118
  14. 2023Do Language Models Know When They're Hallucinating References?cited by 115
  15. 2022ProsocialDialog: A Prosocial Backbone for Conversational Agentscited by 109
  16. 2022Few-Shot Preference Learning for Human-in-the-Loop RLcited by 85
  17. 2023Bactrian-X: Multilingual Replicable Instruction-Following Models with Low-Rank Adaptationcited by 79
  18. 2021Ethical-Advice Taker: Do Language Models Understand Natural Language Interventions?cited by 46
  19. 2023The Trickle-down Impact of Reward (In-)consistency on RLHFcited by 29
  20. 2021On the Multilingual Capabilities of Very Large-Scale English Language Modelscited by 22

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 23; 2025: 34; 2026: 35.

92 papers.

Sort:

By area:

  1. 2024Improving Alignment and Robustness with Circuit Breakerscited by 302
  2. 2024Weak-to-Strong Jailbreaking on Large Language Modelscited by 100
  3. 2024Coercing LLMs to do and reveal (almost) anythingcited by 88
  4. 2024The Multilingual Alignment Prism: Aligning Global and Local Preferences to Reduce Harmcited by 67
  5. 2024JailbreakZoo: Survey, Landscapes, and Horizons in Jailbreaking Large Language and Vision-Language Modelscited by 67
  6. 2024SelfDefend: LLMs Can Defend Themselves against Jailbreaking in a Practical Mannercited by 57
  7. 2025The State of Multilingual LLM Safety Research: From Measuring the Language Gap to Mitigating Itcited by 42
  8. 2024Language Ranker: A Metric for Quantifying LLM Performance Across High and Low-Resource Languagescited by 38
  9. 2025Refusal Direction is Universal Across Safety-Aligned Languagescited by 37
  10. 2024MBBQ: A Dataset for Cross-Lingual Comparison of Stereotypes in Generative LLMscited by 32
  11. 2024How (un)ethical are instruction-centric responses of LLMs? Unveiling the vulnerabilities of safety guardrails to harmful queriescited by 26
  12. 2024Reuse Your Rewards: Reward Model Transfer for Zero-Shot Cross-Lingual Alignmentcited by 26
38 more
  1. 2024Exploring Multilingual Concepts of Human Value in Large Language Models: Is Value Alignment Consistent, Transferable and Controllable across Languages?cited by 23
  2. 2024Adapting LLMs for Efficient Context Processing through Soft Prompt Compressioncited by 22
  3. 2024Multilingual Blending: LLM Safety Alignment Evaluation with Language Mixturecited by 22
  4. 2024From One to Many: Expanding the Scope of Toxicity Mitigation in Language Modelscited by 18
  5. 2024Multilingual Large Language Models: A Systematic Surveycited by 18
  6. 2025Enhancing LLM Safety Through a Theoretical Minimax Game Lenscited by 18
  7. 2025MPO: Multilingual Safety Alignment via Reward Gap Optimizationcited by 18
  8. 2024Mix Data or Merge Models? Optimizing for Diverse Multi-Task Learningcited by 17
  9. 2024Arabic Dataset for LLM Safeguard Evaluationcited by 17
  10. 2025SEA-HELM: Southeast Asian Holistic Evaluation of Language Modelscited by 17
  11. 2025SoK: Evaluating Jailbreak Guardrails for Large Language Modelscited by 17
  12. 2024Jigsaw Puzzles: Splitting Harmful Questions to Jailbreak Large Language Modelscited by 15
  13. 2025Speak Easy: Eliciting Harmful Jailbreaks from LLMs with Simple Interactionscited by 14
  14. 2025Consistency in Language Models: Current Landscape, Challenges, and Future Directionscited by 14
  15. 2026Tiny Aya: Bridging Scale and Multilingual Depthcited by 14
  16. 2024Verifiable by Design: Aligning Language Models to Quote from Pre-Training Datacited by 13
  17. 2025Chain-of-Thought Hijackingcited by 13
  18. 2024Learn and Unlearn: Addressing Misinformation in Multilingual LLMscited by 11
  19. 2025The Multilingual Divide and Its Impact on Global AI Safetycited by 10
  20. 2025The Alignment Waltz: Jointly Training Agents to Collaborate for Safetycited by 10
  21. 2026Obscure but Effective: Classical Chinese Jailbreak Prompt Optimization via Bio-Inspired Searchcited by 10
  22. 2024LLMs Can Defend Themselves Against Jailbreaking in a Practical Manner: A Vision Papercited by 9
  23. 2025The Translation Barrier Hypothesis: Multilingual Generation with Large Language Models Suffers from Implicit Translation Failurecited by 9
  24. 2026Align Once, Benefit Multilingually: Enforcing Multilingual Consistency for LLM Safety Alignmentcited by 9
  25. 2025FinTrust: A Comprehensive Benchmark of Trustworthiness Evaluation in Finance Domaincited by 8
  26. 2025The State of Large Language Models for African Languages: Progress and Challengescited by 7
  27. 2025LionGuard 2: Building Lightweight, Data-Efficient & Localised Multilingual Content Moderatorscited by 7
  28. 2025Linguistic Nepotism: Trading-off Quality for Language Preference in Multilingual RAGcited by 7
  29. 2025Trustworthy Machine Learning via Memorization and the Granular Long-Tail: A Survey on Interactions, Tradeoffs, and Beyondcited by 6
  30. 2025ParaPO: Aligning Language Models to Reduce Verbatim Reproduction of Pre-training Datacited by 6
  31. 2024It Takes Two: On the Seamlessness between Reward and Policy Model in RLHFcited by 5
  32. 2025Adaptive LoRA Merge with Parameter Pruning for Low-Resource Generationcited by 5
  33. 2025It's All About In-Context Learning! Teaching Extremely Low-Resource Languages to LLMscited by 5
  34. 2025Large Multimodal Models for Low-Resource Languages: A Surveycited by 4
  35. 2025Response-Based Knowledge Distillation for Multilingual Jailbreak Prevention Unwittingly Compromises Safetycited by 4
  36. 2025Mind the Gap! Choice Independence in Using Multilingual LLMs for Persuasive Co-Writing Tasks in Different Languagescited by 3
  37. 2025STShield: Single-Token Sentinel for Real-Time Jailbreak Detection in Large Language Modelscited by 3
  38. 2025Optimized Text Embedding Models and Benchmarks for Amharic Passage Retrievalcited by 3

50 most cited of 92 All 92 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025The Hidden Space of Safety: Understanding Preference-Tuned LLMs in Multilingual context
  2. 2026IndicSafe: A Benchmark for Evaluating Multilingual LLM Safety in South Asia
  3. 2026BabelSteering: Multilingual Safety Alignment via English Steering Vectors
  4. 2024Towards Understanding the Fragility of Multilingual LLMs against Fine-Tuning Attacks
  5. 2026Multilingual Refusal Alignment for Safer Large Language Models
  6. 2023All Languages Matter: On the Multilingual Safety of Large Language Models
  7. 2026When Safety Speaks a Language: A Mechanistic Analysis of Safety-Language Identity Entanglement in LLMs
  8. 2025Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges