mlsanityData to 14 Oct 2026

The Hidden Space of Safety: Understanding Preference-Tuned LLMs in Multilingual context

arXiv 2504.02708 ↗First posted 3 Apr 2025cs.CLcited by 6

On the map: LLM reasoning → Reasoning benchmarks and evaluation

also in 1 more
Abstract (arXiv)

Alignment tuning has enabled large language models to excel in reasoning, instruction-following, and minimizing harmful generations. However, despite their widespread deployment, these models exhibit a monolingual bias, raising concerns about the effectiveness of alignment across languages. Current alignment methods predominantly focus on English, leaving it unclear how alignment mechanism generalize to multilingual settings. To address this, we conduct a systematic analysis of distributional shifts in the embedding space of LLMs before and after alignment, uncovering its impact on model behavior across diverse languages. We leverage the alignment-induced separation in safety space as a quantitative tool to measure how alignment enforces safety constraints. Our study evaluates seven LLMs using balanced toxicity datasets and parallel text-detoxification benchmarks, revealing substantial disparities in the latent representation space between high-resource and low-resource languages. These findings underscore the need for language-specific fine-tuning to ensure fair, reliable and robust multilingual alignment. Our insights provide a foundation for developing truly safe multilingual LLMs, emphasizing the urgency of addressing alignment gaps in underrepresented languages.

Key facts. Cited by 6 papers in the map (data through 14 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 21 of its 36 references that are in the map. First posted on arXiv 3 Apr 2025.

Builds on

21 of 36 references.

References · 21

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2017Proximal Policy Optimization Algorithmscited by 15,215
  3. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  4. 2017Deep reinforcement learning from human preferencescited by 3,912
  5. 2020Learning to summarize from human feedbackcited by 2,588
  6. 2019Fine-Tuning Language Models from Human Preferencescited by 2,150
  7. 2023A General Theoretical Paradigm to Understand Learning from Human Preferencescited by 912
  8. 2023Jailbreaking ChatGPT via Prompt Engineering: An Empirical Studycited by 498
  9. 2023Scalable Extraction of Training Data from (Production) Language Modelscited by 434
  10. 2023Understanding the Effects of RLHF on LLM Generalisation and Diversitycited by 396
  11. 2023Analyzing Leakage of Personally Identifiable Information in Language Modelscited by 285
  12. 2023LaMini-LM: A Diverse Herd of Distilled Models from Large-Scale Instructionscited by 124
9 more
  1. 2024RLHF Deciphered: A Critical Analysis of Reinforcement Learning from Human Feedback for LLMscited by 90
  2. 2024Towards Understanding Jailbreak Attacks in LLMs: A Representation Space Analysiscited by 64
  3. 2024Binary Classifier Optimization for Large Language Model Alignmentcited by 45
  4. 2024RLHF Can Speak Many Languages: Unlocking Multilingual Preference Optimization for LLMscited by 43
  5. 2025Multilingual != Multicultural: Evaluating Gaps Between Multilingual Capabilities and Cultural Alignment in LLMscited by 39
  6. 2024MM-Eval: A Multilingual Meta-Evaluation Benchmark for LLM-as-a-Judge and Reward Modelscited by 35
  7. 2023On Evaluating and Mitigating Gender Biases in Multilingual Settingscited by 31
  8. 2024Preference Tuning with Human Feedback on Language, Speech, and Vision Tasks: A Surveycited by 23
  9. 2025LLM Safety Alignment is Divergence Estimation in Disguisecited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 3; 2026: 3.

6 papers.

  1. 2025Refusal Direction is Universal Across Safety-Aligned Languagescited by 38
  2. 2025MPO: Multilingual Safety Alignment via Reward Gap Optimizationcited by 18
  3. 2025Are LLMs Empathetic to All? Investigating the Influence of Multi-Demographic Personas on a Model's Empathycited by 10
  4. 2026LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safetycited by 3
  5. 2026Low-Resource Safety Failures Are Action Failures, Not Representation Failurescited by 2
  6. 2026LLM Safety Alignment in Low-Resource Languages: A Systematic Literature Reviewcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts
  2. 2026Align Once, Benefit Multilingually: Enforcing Multilingual Consistency for LLM Safety Alignment
  3. 2024Multilingual Blending: LLM Safety Alignment Evaluation with Language Mixture
  4. 2025Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges
  5. 2025Picky LLMs and Unreliable RMs: An Empirical Study on Safety Alignment after Instruction Tuning
  6. 2025A Comprehensive Evaluation framework of Alignment Techniques for LLMs
  7. 2025Safety Alignment Depth in Large Language Models: A Markov Chain Perspective
  8. 2026Multilingual Refusal Alignment for Safer Large Language Models