Don't Want Your LLM to Recommend Nuclear Strike? Try Asking It in Japanese

Rian Touchent

Large language models are increasingly used in strategic and advisory contexts, yet their safety alignment is typically evaluated in English only. We test nine models from six prov…