mlsanityData to 14 Oct 2026

The Multilingual Alignment Prism: Aligning Global and Local Preferences to Reduce Harm

arXiv 2406.18682 ↗First posted 26 Jun 2024cs.CL cs.AI cs.LGcited by 70

On the map: LLM safety and security → Safety alignment, refusal and guardrails

Abstract (arXiv)

A key concern with the concept of "alignment" is the implicit question of "alignment to what?". AI systems are increasingly used across the world, yet safety alignment is often focused on homogeneous monolingual settings. Additionally, preference training and safety measures often overfit to harms common in Western-centric datasets. Here, we explore the viability of different alignment approaches when balancing dual objectives: addressing and optimizing for a non-homogeneous set of languages and cultural preferences while minimizing both global and local harms. We collect the first set of human annotated red-teaming prompts in different languages distinguishing between global and local harm, which serve as a laboratory for understanding the reliability of alignment techniques when faced with preference distributions that are non-stationary across geographies and languages. While this setting is seldom covered by the literature to date, which primarily centers on English harm mitigation, it captures real-world interactions with AI systems around the world. We establish a new precedent for state-of-the-art alignment techniques across 6 languages with minimal degradation in general performance. Our work provides important insights into cross-lingual transfer and novel optimization approaches to safeguard AI systems designed to serve global populations.

Key facts. Cited by 70 papers in the map (data through 14 Oct 2026). Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security. Builds on 79 of its 110 references that are in the map. First posted on arXiv 26 Jun 2024.

Builds on

79 of 110 references.

Closest ancestors · 3

Same direction, newest first

  1. 2024Safety Alignment Should Be Made More Than Just a Few Tokens Deep10 Jun
  2. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!5 Oct
  3. 2023All Languages Matter: On the Multilingual Safety of Large Language Models2 Oct

Other references · 76

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2023GPT-4 Technical Reportcited by 18,396
  3. 2022Training language models to follow instructions with human feedbackcited by 15,971
  4. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  5. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  6. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  7. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  8. 2017Deep reinforcement learning from human preferencescited by 3,912
  9. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,780
  10. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,720
  11. 2022Constitutional AI: Harmlessness from AI Feedbackcited by 3,050
  12. 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 2,724
38 more
  1. 2020Learning to summarize from human feedbackcited by 2,588
  2. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,651
  3. 2022No Language Left Behind: Scaling Human-Centered Machine Translationcited by 1,312
  4. 2020RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Modelscited by 1,255
  5. 2023WizardLM: Empowering large pre-trained language models to follow complex instructionscited by 1,125
  6. 2022Red Teaming Language Models with Language Modelscited by 1,026
  7. 2023WizardCoder: Empowering Code Large Language Models with Evol-Instructcited by 733
  8. 2020Aligning AI With Shared Human Valuescited by 723
  9. 2019Universal Adversarial Triggers for Attacking and Analyzing NLPcited by 722
  10. 2022Automatic Chain of Thought Prompting in Large Language Modelscited by 688
  11. 2023Safe RLHF: Safe Reinforcement Learning from Human Feedbackcited by 656
  12. 2023Bias and Fairness in Large Language Models: A Surveycited by 639
  13. 2019The Woman Worked as a Babysitter: On Biases in Language Generationcited by 541
  14. 2017Ray: A Distributed Framework for Emerging AI Applicationscited by 527
  15. 2023Zephyr: Direct Distillation of LM Alignmentcited by 506
  16. 2023Jailbreaking ChatGPT via Prompt Engineering: An Empirical Studycited by 498
  17. 2018Scalable agent alignment via reward modeling: a research directioncited by 447
  18. 2023Scalable Extraction of Training Data from (Production) Language Modelscited by 434
  19. 2023Is ChatGPT a Good NLG Evaluator? A Preliminary Studycited by 422
  20. 2023Large Language Models Are State-of-the-Art Evaluators of Translation Qualitycited by 420
  21. 2023Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructionscited by 387
  22. 2023Toxicity in ChatGPT: Analyzing Persona-assigned Language Modelscited by 360
  23. 2023Are aligned neural networks adversarially aligned?cited by 315
  24. 2024Aya Model: An Instruction Finetuned Open-Access Multilingual Language Modelcited by 295
  25. 2023Analyzing Leakage of Personally Identifiable Information in Language Modelscited by 285
  26. 2023Multilingual Jailbreak Challenges in Large Language Modelscited by 240
  27. 2022Challenges and Strategies in Cross-Cultural NLPcited by 218
  28. 2022Predictability and Surprise in Large Generative Modelscited by 210
  29. 2023Okapi: Instruction-tuned Large Language Models in Multiple Languages with Reinforcement Learning from Human Feedbackcited by 190
  30. 2023Gender bias and stereotypes in Large Language Modelscited by 187
  31. 2024Preference Fine-Tuning of LLMs Should Leverage Suboptimal, On-Policy Datacited by 185
  32. 2024Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuningcited by 177
  33. 2023MART: Improving LLM Safety with Multi-round Automatic Red-Teamingcited by 163
  34. 2022"I'm sorry to hear that": Finding New Biases in Language Models with a Holistic Descriptor Datasetcited by 162
  35. 2022Probing Pre-Trained Language Models for Cross-Cultural Differences in Valuescited by 149
  36. 2024The PRISM Alignment Dataset: What Participatory, Representative and Individualised Human Feedback Reveals About the Subjective and Multicultural Alignment of Large Language Modelscited by 144
  37. 2024Collective Constitutional AI: Aligning a Language Model with Public Inputcited by 123
  38. 2021Gender Bias in Machine Translationcited by 117

50 most cited of 76 All 76 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 21; 2025: 28; 2026: 21.

70 papers.

Sort:

By area:

  1. 2025Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMscited by 172
  2. 2024Aya Expanse: Combining Research Breakthroughs for a New Multilingual Frontiercited by 129
  3. 2025Large Language Models Often Know When They Are Being Evaluatedcited by 128
  4. 2024INCLUDE: Evaluating Multilingual Language Understanding with Regional Knowledgecited by 86
  5. 2024Large Language Model Safety: A Holistic Surveycited by 59
  6. 2025PolyGuard: A Multilingual Safety Moderation Tool for 17 Languagescited by 59
  7. 2025Detecting High-Stakes Interactions with Activation Probescited by 53
  8. 2024M-RewardBench: Evaluating Reward Models in Multilingual Settingscited by 50
  9. 2024To Code, or Not To Code? Exploring Impact of Code in Pre-trainingcited by 47
  10. 2025The State of Multilingual LLM Safety Research: From Measuring the Language Gap to Mitigating Itcited by 45
  11. 2024Controllable Safety Alignment: Inference-Time Adaptation to Diverse Safety Requirementscited by 36
  12. 2024Towards Understanding the Fragility of Multilingual LLMs against Fine-Tuning Attackscited by 33
38 more
  1. 2025CultureGuard: Towards Culturally-Aware Dataset and Guard Model for Multilingual Safety Applicationscited by 27
  2. 2025Safer or Luckier? LLMs as Safety Evaluators Are Not Robust to Artifactscited by 26
  3. 2024Preference Tuning with Human Feedback on Language, Speech, and Vision Tasks: A Surveycited by 23
  4. 2025Enhancing LLM Safety Through a Theoretical Minimax Game Lenscited by 19
  5. 2024On the Limitations of Compute Thresholds as a Governance Strategycited by 18
  6. 2025MPO: Multilingual Safety Alignment via Reward Gap Optimizationcited by 18
  7. 2024WalledEval: A Comprehensive Safety Evaluation Toolkit for Large Language Modelscited by 17
  8. 2024Mix Data or Merge Models? Optimizing for Diverse Multi-Task Learningcited by 17
  9. 2024EMMA-500: Enhancing Massively Multilingual Adaptation of Large Language Modelscited by 16
  10. 2025CARE: Multilingual Human Preference Learning for Cultural Awarenesscited by 15
  11. 2026Tiny Aya: Bridging Scale and Multilingual Depthcited by 15
  12. 2024The Future of Open Human Feedbackcited by 10
  13. 2024Insights and Current Gaps in Open-Source LLM Vulnerability Scanners: A Comparative Analysiscited by 10
  14. 2025The Multilingual Divide and Its Impact on Global AI Safetycited by 10
  15. 2025Implicit Cross-Lingual Rewarding for Efficient Multilingual Preference Alignmentcited by 9
  16. 2025MrGuard: A Multilingual Reasoning Guardrail for Universal LLM Safetycited by 9
  17. 2025When Life Gives You Samples: The Benefits of Scaling up Inference Compute for Multilingual LLMscited by 9
  18. 2025Societal Alignment Frameworks Can Improve LLM Alignmentcited by 8
  19. 2026MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM Safetycited by 8
  20. 2024Multilingual Arbitrage: Optimizing Data Pools to Accelerate Multilingual Progresscited by 7
  21. 2024LASeR: Learning to Adaptively Select Reward Models with Multi-Armed Banditscited by 7
  22. 2024LLMs Lost in Translation: M-ALERT uncovers Cross-Linguistic Safety Inconsistenciescited by 7
  23. 2025MAPS: A Multilingual Benchmark for Agent Performance and Securitycited by 7
  24. 2025OMNIGUARD: An Efficient Approach for AI Safety Moderation Across Languages and Modalitiescited by 7
  25. 2025When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Modelscited by 6
  26. 2026Translation as a Scalable Proxy for Multilingual Evaluationcited by 6
  27. 2024Cross-Lingual Transfer of Debiasing and Detoxification in Multilingual LLMs: An Extensive Investigationcited by 5
  28. 2025TRIDENT: Enhancing Large Language Model Safety with Tri-Dimensional Diversified Red-Teaming Data Synthesiscited by 4
  29. 2025Massively Multilingual Adaptation of Large Language Models Using Bilingual Translation Datacited by 4
  30. 2025M2PO: Multi-Perspective Multi-Pair Preference Optimization for Machine Translationcited by 4
  31. 2026AI Alignment Breaks at the Edgecited by 4
  32. 2024With a Grain of SALT: Are LLMs Fair Across Social Dimensions?cited by 3
  33. 2025CAPO: Confidence Aware Preference Optimization Learning for Multilingual Preferencescited by 3
  34. 2025K2-V2: A 360-Open, Reasoning-Enhanced LLMcited by 3
  35. 2024Towards Safe Multilingual Frontier AIcited by 2
  36. 2025Social Bias in Multilingual Language Models: A Surveycited by 2
  37. 2025CREST: Universal Safety Guardrails Through Cluster-Guided Cross-Lingual Transfercited by 2
  38. 2025AprielGuardcited by 2

50 most cited of 70 All 70 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Bridging the Multilingual Safety Divide: Efficient, Culturally-Aware Alignment for Global South Languages
  2. 2024Unintended Impacts of LLM Alignment on Global Representation
  3. 2025Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges
  4. 2025The Hidden Space of Safety: Understanding Preference-Tuned LLMs in Multilingual context
  5. 2026The Illusion of Cross-Lingual Safety in Low-Resource Languages
  6. 2025Cultivating Pluralism In Algorithmic Monoculture: The Community Alignment Dataset
  7. 2024Reuse Your Rewards: Reward Model Transfer for Zero-Shot Cross-Lingual Alignment
  8. 2024SafeWorld: Geo-Diverse Safety Alignment