Beyond English: Gauging AI Safety in South Asian Languages

A new benchmark reveals that current methods for evaluating AI vulnerability are often misleading when applied to languages beyond English, particularly in the diverse linguistic landscape of South Asia.






