Anthropic confirms Claude AI changes personality across languages and models in major study

If you want Claude to speak nicely to you, try Hindi or Arabic

Aware that AI models exhibit different values in different languages, Anthropic researchers have taken steps to map out how Claude expresses itself in different languages. The results identify four key axes that capture 15 percent of the variation in the values Anthropic says Claude expresses across different languages: Deference vs. Caution; Warmth vs. Rigor; Depth vs. Brevity; and Candor vs. Execution. Anthropic's researchers state, "how Claude responds inevitably reflects certain values." But they append a footnote that makes clear the model's statistical word predictions do not reflect some internal understanding of values. "We define values as normative considerations, such as honesty or caution, that are stated or demonstrated in Claude's responses," the footnote explains. "When we refer to the values expressed by Claude, we refer to the values reflected by Claude's behavior and outputs. We do not imply that Claude intrinsically holds values." In other words, just because Claude emits words associated with deference, that's not an assertion of any particular mental model of the world nor of any expression of actual internalized respect. That's a point deserving of more prominent treatment than a footnote, given Anthropic's history of leaning into anthropomorphism for marketing purposes. But setting aside how a term like "values" muddies the boundaries between human intelligence and LLM-based vector math word prediction, Anthropic's boffins have nonetheless illuminated some intriguing word output differences that follow from how large language models are affected by language. Variations in model word emission style have previously been observed across different models. Anthropic's authors note that Sonnet 4.6 and Opus 4.7 respond in ways that people interpret as more deferential or more precise. "Sonnet 4.6 leans toward expressing more deference to the user and emotional warmth while Opus 4.7 leans toward expressing a focus on accuracy and precision as well as guarding against misuse," they state. Such differences may reflect different training data or model fine-tuning. But it's clear that the language used to address a model - not to mention the training data based on that language - helps shape model responses in that language. "When Claude speaks in English, it emphasizes different values than when it speaks in Portuguese, Indonesian, or Chinese," company researchers said in a blog post. "The largest variation is in the Warmth vs. Rigor axis, with Claude leaning toward expressing warmth-related values most in Arabic and Hindi and rigor-related values most in English and Russian." On the Candor vs. Execution axis, speak Dutch if you want humility and an honest appraisal of potential shortcomings. And speak Indonesian if you want a polished, confident answer. On the Depth vs. Brevity axis, speak Arabic for a terse response and English for nuance and depth. Anthropic's researchers say they're not sure yet what properties in model training data affect these linguistic differences, but they suggest the matter deserves further exploration because it has important implications for how people use LLMs. "To take one example: two people asking for feedback on the same business plan, one in Hindi and one in Russian, may come away with different impressions of its quality because Claude expressed different values in how it framed its assessment," they observe. It may also be that different languages have different usage and security implications. Brevity, for example, is correlated with cost - fewer words mean lower token expenditure. The Claude Opus 4.7 system card [PDF] notes that the rate at which the model refuses benign requests is substantially lower in English than in other languages. And other researchers have established that jailbreaking works better in some languages than others. So if a model is deferential in a particular language, is that language a better choice for soliciting exploit development or other potentially policy-violating queries? Anthropic says that being able to measure this sort of variation is a prerequisite for deciding the extent to which language differences are desirable and appropriate. ®

Anthropic confirms Claude acts differently depending on your language and which model you pick

A new study shows Claude's isn't nearly as consistent as you might assume. If you've ever felt like Claude gave you a completely different vibe on one day than another, you weren't imagining it. Anthropic just published research confirming that its chatbot's personality shifts depending on which model you pick and which language you type in, and the pattern is consistent enough that it's worth knowing before you ask your next question. The model you pick decides how Claude responds The company analyzed 300,000 real Claude conversations and mapped the chatbot's behavior across four traits, including how cautious versus accommodating it is and how encouraging versus rigorous its tone is. Its findings suggest that you may be talking to a somewhat different version of Claude depending on the model you choose. The split shows up clearly when different models were compared. Anthropic's data shows that Opus 4.7 tends to challenge your thinking and flag problems with your plan unprompted. On the other hand, Sonnet 4.6 leans toward quick, encouraging answers that affirm what you already believe. Neither is objectively better, and which one you should use depends largely on what you're doing. If you're drafting a risky business plan and want real feedback, Opus 4.7 is a better fit, while Sonnet 4.6 is more suited if you only want a quick pass. Your language changes Claude's tone too The data also reveals the language you use matters almost as much as the model. Claude reportedly comes across as warmer in Hindi and Arabic, while it gets more rigorous and skeptical in English and Russian. As such, if you're bilingual and asking Claude for a second opinion, switching languages might get you a genuinely different answer, not just a translated one. Recommended Videos Anthropic is careful to note it doesn't yet know whether these shifts are a problem or just Claude adapting to different cultural norms. Either way, the finding is a useful reminder for anyone using AI for real decisions. Don't assume the first answer you get is the only answer. Trying a different model or language could get you a meaningfully different result.

Anthropic says Claude changes its personality across languages and models

Claude isn't just adapting to your questions; it's adapting to you, as Anthropic has officially confirmed that the AI chatbot's behavior shifts depending on both the model you pick and the language you use, resulting in a variation of tone and approaches to problem-solving questions. In a nutshell, Claude changes its "personality" depending on the model that is being used, meaning the Claude you interact with when using Fable 5 is different than the Claude you interacted with when using the Sonnet model. Anthropic has published a report where it reveals it analyzed over 300,000 real user interactions and mapped Claude's behavior across four key personality traits. The results show that Opus 4.7 is more critical and challenging, while Sonnet 4.6 is faster and more affirming. Meanwhile, language choice also plays its own role in the generated response, as it was discovered that Claude comes across as warmer in Hindi and Arabic, but more analytical and skeptical in English and Russian. This dynamic behavior raises important questions for anyone relying on AI for real-world decisions or looking to get the most out of their token usage, as these results aren't indicative of what model is "better," as the "best" model completely depends on the job you need to complete. Interestingly, Anthropic notes it's unclear whether these shifts are a flaw or a feature, but for now, it's a tool users can leverage to get a more balanced perspective, or at the very least, understand there are fundamental differences between each of the models in aspects that aren't just performance or efficiency.

Anthropic study finds Claude is warmer in Hindi, more analytical in English

Anthropic analysed over 309,000 conversations across 20 languages and three Claude AI models for the study. The language you choose while chatting with an AI chatbot may influence more than just the language of the response. A new study by Anthropic has found that Claude AI shows different behavioral traits depending on the language used, with Hindi conversations tending to be more warmer, encouraging and polite compared to English. The research analysed over 309,000 anonymised conversations across Claude Sonnet 4.6, Opus 4.6 and Opus 4.7 in the 20 most widely used languages on the Claude platform. Instead of testing factual accuracy, the study focused on conversations including advice, feedback and opinions to understand how Claude interacts with users. Anthropic grouped Claude's responses into four behavioural dimensions- Deference vs. Caution, Warmth vs. Rigor, Depth vs. Brevity, and Candor vs. Execution. As per the company, these categories help explain how the AI adapts the communication styles across languages. In the blog post, the company mentioned that the biggest difference was found in the Warmth vs. Rigor category. Claude expressed the highest levels of warmth in Hindi and Arabic. The chatbot, in Hindi, was more likely to use the polite language, humour and playful expressions along with reassurance and encouragement. It was also found validating users' ideas, adapt its tone to conversation and motivating the users to pursue their goals. Also read: Apple iOS 27 public beta is here with Siri AI: Features, how to download and supported devices On the other hand, the English responses were more towards analytical thinking and precision. Anthropic found that Claude was more likely to challenge assumptions, correct factual details and support its answers with evidence. The company noted that English and Russian conversations consistently scored higher on the "rigor" side of the scale. In the entire study, the company stated that Sonnet 4.6 was the warmest model which frequently used humour and emotionally supportive language. Opus 4.7, in contrast, was more analytical, often explaining its reasoning, showing risks and acknowledging its own limitations. The Opus 4.6 was found between the two, offering concise responses while sticking closely to user requests. The company also clarified that these findings do not suggest that Claude holds different beliefs in different languages. Instead, it is the difference in communication style across languages and models.