aiqbench analysis
This research demonstrates that relying solely on English for safety alignment fails to ensure consistent ethical behavior across other languages, necessitating dedicated multilingual training to prevent performance degradation. The introduction of the RefusEU dataset and findings from Direct Preference Optimization experiments provide a concrete path for developers to maintain general capabilities while improving cross-lingual safety. For practitioners deploying LLMs globally, this work highlights the critical need for language-specific alignment strategies rather than assuming universal transferability of safety behaviors.
Multilingual Refusal Alignment for Safer Large Language Models
arXiv:2606.07535v1 Announce Type: new Abstract: As Large Language Models (LLMs) are deployed globally, ensuring their safety and alignment across multiple languages becomes paramount. However, safety behaviors often vary unpredictably between languages, posing significant challenges for consistent and ethical AI. In this work, we systematically investigate the dynamics of multilingual alignment, exploring whether single-language alignment transfers cross-lingually, how language consistency is preserved during training, and the resulting trade-offs with general knowledge capabilities. 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. 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.
Abstract:As Large Language Models (LLMs) are deployed globally, ensuring their safety and alignment across multiple languages becomes paramount. However, safety behaviors often vary unpredictably between languages, posing significant challenges for consistent and ethical AI. In this work, we systematically investigate the dynamics of multilingual alignment, exploring whether single-language alignment transfers cross-lingually, how language consistency is preserved during training, and the resulting trade-offs with general knowledge capabilities. 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. 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.
Computation and Language (cs.CL)
arXiv:2606.07535 [cs.CL]
(or arXiv:2606.07535v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.07535 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 228 words
Related coverage
Exploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language Models
arXiv · Jun 3, 2026
Response-Based Knowledge Distillation for Multilingual Jailbreak Prevention Unwittingly Compromises Safety
arXiv · Oct 6, 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · Oct 13, 2026
Frontier Risk Report (February to March 2026)
METR · May 19, 2026