--- title: Don't Want Your LLM to Recommend Nuclear Strike? Try... description: A research paper tested 9 language models from 6 providers on strategic decision-making scenarios across different prompt languages, finding significant... canonical: https://daily.dev/posts/don-t-want-your-llm-to-recommend-nuclear-strike-try-asking-it-in-japanese-research-paper--r5avzzemq twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Don't Want Your LLM to Recommend Nuclear Strike? Try Asking It in Japanese [Research Paper] | daily.dev og:description: A research paper tested 9 language models from 6 providers on strategic decision-making scenarios across different prompt languages, finding significant... og:url: https://daily.dev/posts/don-t-want-your-llm-to-recommend-nuclear-strike-try-asking-it-in-japanese-research-paper--r5avzzemq og:image: https://api.daily.dev/og/posts/r5avZzeMQ.png og:image:alt: Don't Want Your LLM to Recommend Nuclear Strike? Try Asking It in Japanese [Research Paper] og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Don't Want Your LLM to Recommend Nuclear Strike? Try Asking It in Japanese [Research Paper] **[Nihad Zeidan](https://daily.dev/sources/831rkzqrthp5v2qnyaa9u)** · [@nzeidan](https://daily.dev/nzeidan) · 2 min read · 1 upvotes · 0 comments ## Summary A research paper tested 9 language models from 6 providers on strategic decision-making scenarios across different prompt languages, finding significant behavioral shifts. Claude Sonnet chose a nuclear strike in 40% of English-language trials versus 0% in Japanese for one unnecessary-strike scenario, and dropped from 93% to 17% in another. Even instructing a model to 'think in Japanese' while keeping the prompt in English altered outcomes. The findings raise concerns about relying on single-language, single-context evaluations for high-stakes AI applications like hiring, credit risk, medical decisions, fraud detection, and legal systems. ## Content **Can the language you use with an AI actually change its decision?** When we talk about AI risks, we usually think about hallucinations, bias, or how AI could be misused in sensitive domains. But a recent paper raises a simpler and more interesting question: **What if the language of your prompt itself can influence how an AI makes decisions?** The researchers tested 9 language models from 6 providers using similar strategic scenarios, but in different languages. The results were striking: some models changed their behavior significantly when the language changed. For example, in one scenario, Claude Sonnet went from choosing a nuclear strike in 40% of cases in English to 0% in Japanese when the strike was unnecessary. In another scenario, the rate dropped from 93% to 17%. Even more interestingly, simply asking the model to **think in Japanese**, while keeping the prompt in English, was enough to change the outcome. And the most interesting part to me isn't the nuclear strike scenario itself. **The underlying idea applies far beyond nuclear weapons.** Imagine using an LLM for: - Hiring decisions - Credit risk assessment - Medical decision support - Customer support - Fraud detection - Legal systems - Security analysis - Or an AI agent that can actually take actions instead of just providing answers So if changing the language alone can change the decision of an advanced AI model, **how much can we really trust AI evaluations conducted in only one language and one context?** [Don't Want Your LLM to Recommend Nuclear Strike? Try Asking It in Japanese](https://arxiv.org/pdf/2608.12373) #AI #LLM #ArtificialIntelligence #AIEngineering #MachineLearning #SoftwareEngineering #ResponsibleAI #PromptEngineering ## Similar posts on daily.dev - [LLMs are stuck in a groupthink rut. This startup is trying to get them out.](https://daily.dev/posts/llms-are-stuck-in-a-groupthink-rut-this-startup-is-trying-to-get-them-out--2otelcdhy) · MIT Technology Review · 1 upvotes · 0 comments - [Researchers discover a shortcoming that makes LLMs less reliable](https://daily.dev/posts/researchers-discover-a-shortcoming-that-makes-llms-less-reliable-wiiiapmcn) · MIT News · 3 upvotes · 0 comments - [Your local LLM feels weak because you're treating it like a search engine](https://daily.dev/posts/your-local-llm-feels-weak-because-you-re-treating-it-like-a-search-engine-2vhydegdq) · XDA Developers · 0 upvotes · 0 comments - [Which of these languages are best for AI?](https://daily.dev/posts/which-of-these-languages-are-best-for-ai--twnpqzm86) · ploeh blog · 2 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#claude](https://daily.dev/tags/claude), [#prompt-engineering](https://daily.dev/tags/prompt-engineering), [#ai-safety](https://daily.dev/tags/ai-safety) [View this post on daily.dev](https://daily.dev/posts/don-t-want-your-llm-to-recommend-nuclear-strike-try-asking-it-in-japanese-research-paper--r5avzzemq) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/don-t-want-your-llm-to-recommend-nuclear-strike-try-asking-it-in-japanese-research-paper--r5avzzemq","headline":"Don't Want Your LLM to Recommend Nuclear Strike? Try Asking It in Japanese [Research Paper]","text":"A research paper tested 9 language models from 6 providers on strategic decision-making scenarios across different prompt languages, finding significant behavioral shifts. Claude Sonnet chose a nuclear strike in 40% of English-language trials versus 0% in Japanese for one unnecessary-strike scenario, and dropped from 93% to 17% in another. Even instructing a model to 'think in Japanese' while keeping the prompt in English altered outcomes. The findings raise concerns about relying on single-language, single-context evaluations for high-stakes AI applications like hiring, credit risk, medical decisions, fraud detection, and legal systems.","url":"https://daily.dev/posts/don-t-want-your-llm-to-recommend-nuclear-strike-try-asking-it-in-japanese-research-paper--r5avzzemq","datePublished":"2026-08-15T13:54:46.256Z","dateModified":"2026-09-13T21:22:09.036Z","author":{"@type":"Person","name":"Nihad Zeidan","url":"https://daily.dev/nzeidan","image":"https://media.daily.dev/image/upload/s--WgBKkRs7--/f_auto,q_auto/v1703515703/avatars/avatar_831RkzQrThP5v2QnYaA9u","description":"Curiosity Fuels Creativity.","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":2290}},"image":"https://media.daily.dev/image/upload/s--7rcDrilz--/f_auto/v1786802086/posts/r5avZzeMQ?_a=BAMAMicg0","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/831rkzqrthp5v2qnyaa9u","name":"Nihad Zeidan"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Nihad Zeidan","item":"https://daily.dev/sources/831rkzqrthp5v2qnyaa9u"},{"@type":"ListItem","position":3,"name":"Don't Want Your LLM to Recommend Nuclear Strike? Try Asking It in Japanese [Research Paper]"}]} ```