Which Institutional Frameworks Do Chatbots Assume? Auditing Jurisdictional Defaults in Multilingual LLMs

In three linesAudit of 7 LLMs (US/China) on 2,520 responses to 60 legal-administrative prompts in English and Mandarin. Models default to the institutional framework of input language: 74.5% of English responses adopt US framework, 53.3% of Chinese responses adopt China framework. Risk of jurisdictional misselection when preferred language differs from applicable jurisdiction.

Summary generated by Claude — human-verified