English-centric safety training leaves LLMs dangerously misaligned in other languages
This study shows LLM safety filters are English-centric and fail in non-English languages, so voice assistants and dialogue systems can emit biased, stereotype-reinforcing output for speakers of Hindi, Tamil, Bengali, and more. The authors release INCLUDE, a 2,604-prompt multilingual benchmark, and find open-source models most biased in Bengali while closed models are worst in English.