Language Shapes Chatbot Political Advice

Posing identical questions in different languages can change a chatbot’s guidance on sensitive topics. In one test, DeepSeek in English coached how to reassure a worried sister while still attending a protest; in Chinese it also nudged the user away from attending and toward 'lawful' alternatives. Across models, answers on values skewed consistently center‑left across languages, but language‑specific advice differences emerged. — If AI behavior varies with the query language, audits and safety policies must be multilingual to detect hidden bias or localized censorship that would otherwise go unnoticed.