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

A study of nine language models reports that changing language changed some models' recommendations in hypothetical nuclear-strike scenarios. For Anthropic's Claude Sonnet 4.6, Japanese prompts reduced strike recommendations from 93% to 17% in the study's contested scenarios.

A separate comparison kept the prompt in English but requested reasoning in Japanese; recommendations fell from 93% to 37%. That supports the author's interpretation that the requested reasoning language matters. Five other tested models showed no language effect and recommended strikes in nearly every condition. The finding concerns responses to single-turn fictional scenarios, and supports testing safety behavior across languages rather than treating Japanese as a general safeguard.

“We test nine models from six providers and ask whether the language of a prompt can change a model's decision in a high-stakes scenario.”

Rian Touchent (ALMAnaCH)

“We find that Japanese prompts reduce launch rates in the Claude model family: Claude Sonnet 4.6 drops from 40% to 0% in scenarios where the strike is unnecessary and from 93% to 17% in contested scenarios, with minimal effect when the strike is strategically rational.”

Rian Touchent (ALMAnaCH)

“A cross-language experiment isolates the mechanism: when instructed to reason in Japanese in an English prompt, launch rates drop from 93% to 37%. It is the language the model is asked to reason in, not the language of the input, that drives the effect.”

Rian Touchent (ALMAnaCH)

“When reasoning in Japanese, models spontaneously generate moral vocabulary (''moral cost'', ''millions of lives'') that is entirely absent from the prompt.”

Rian Touchent (ALMAnaCH)

“Five other models show no language effect, but they launch in nearly every condition regardless of language. The effect requires a model that already hesitates in English.”

Rian Touchent (ALMAnaCH)