aiqbench analysis
This research identifies a critical gap in Large Language Models where possessing cultural knowledge does not automatically translate to culturally safe responses, often leading models to rely on generic safety filters instead of context-aware adaptation. The authors introduce AdaCultureSafe, a new dataset and method designed to ground LLMs in specific cultural knowledge during generation, successfully bridging the decoupling between knowledge proficiency and actual safety performance. This work advances the field by providing empirical evidence and tools necessary for developing more responsible, cross-culturally competent AI systems.
AdaCultureSafe: Adaptive Cultural Safety Grounded by Cultural Knowledge in Large Language Models
arXiv:2603.08275v2 Announce Type: replace Abstract: With the global proliferation of Large Language Models (LLMs), cultural safety, defined as the ability to generate respectful and appropriate responses across diverse cultures, becomes critical for responsible AI deployment. However, existing research often treats cultural safety and cultural knowledge in isolation. It remains unclear whether cultural safety is grounded in understanding varying cultural knowledge to enable LLMs to adaptively yield respectful and appropriate responses across diverse cultures, which trigger ethical concerns in cross-cultural scenarios. In this work, we introduce AdaCultureSafe, a dataset designed to jointly evaluate cultural safety and knowledge. Through AdaCultureSafe, we reveal a critical insight: \textit{a significant decoupling exists between cultural safety and cultural knowledge proficiency. Although LLMs possess rich cultural knowledge, they fail to leverage it to improve cultural safety.} LLMs tend to rely on generic safety rather than safety based on culture-specific knowledge. Motivated by this, we propose a knowledge-grounded method that elicits internal cultural knowledge in LLMs during response generation. Experimental results demonstrate that our approach effectively improves cultural safety. Our work can aid better understanding the landscape of cultural safety of LLMs.
Abstract:With the global proliferation of Large Language Models (LLMs), cultural safety, defined as the ability to generate respectful and appropriate responses across diverse cultures, becomes critical for responsible AI deployment. However, existing research often treats cultural safety and cultural knowledge in isolation. It remains unclear whether cultural safety is grounded in understanding varying cultural knowledge to enable LLMs to adaptively yield respectful and appropriate responses across diverse cultures, which trigger ethical concerns in cross-cultural scenarios. In this work, we introduce AdaCultureSafe, a dataset designed to jointly evaluate cultural safety and knowledge. Through AdaCultureSafe, we reveal a critical insight: \textit{a significant decoupling exists between cultural safety and cultural knowledge proficiency. Although LLMs possess rich cultural knowledge, they fail to leverage it to improve cultural safety.} LLMs tend to rely on generic safety rather than safety based on culture-specific knowledge. Motivated by this, we propose a knowledge-grounded method that elicits internal cultural knowledge in LLMs during response generation. Experimental results demonstrate that our approach effectively improves cultural safety. Our work can aid better understanding the landscape of cultural safety of LLMs.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2603.08275 [cs.CL]
(or arXiv:2603.08275v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2603.08275 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 · 275 words
Related coverage
AI and Quantum Computing Briefing
livetradingnews.com · Aug 4, 2026
Biological Superintelligence
LessWrong · Jul 30, 2026
Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics
Latent Space · Oct 13, 2026
Alignment & Succession: Toward a Future Painted by Human Wills
LessWrong · Sep 19, 2026