AI Safety Collapses in Non-English Languages, Misinformation Risk Grows

Original: Sycophancy as a Multilingual Alignment Failure: How Safety Degrades Across Languages, Topics, and Models

Source: arxiv.org ↗