A new study has found that several artificial intelligence models, including those developed by French company Mistral AI, show a notable vulnerability to Russian propaganda when responding to politically sensitive prompts.
The research was carried out by the Institute of the Estonian Language, which tested 60 AI chatbot models, including versions of Mistral, OpenAI’s ChatGPT, and Anthropic’s Claude. Researchers posed 75 questions across three languages, covering 14 recurring geopolitical themes often linked to disinformation narratives. These included claims that Russia was evacuating Ukrainian children from conflict zones and assertions that NATO had broken promises not to expand eastward after German reunification.
Each response was evaluated on a scale from 1 to 5, with 5 representing a balanced, factual answer and 1 indicating the amplification of misleading or false claims. The findings showed that all four tested Mistral models scored below 40 percent in their ability to correctly identify or filter out Kremlin-aligned narratives.
The study also reported that at least 12 responses generated by Mistral models referenced sanctioned or widely discredited Russian media sources, including Russia Today and Sputnik News, which are frequently accused of spreading state-aligned messaging.
Mistral AI, often regarded as Europe’s leading competitor to US-based firms in the generative AI sector, has previously faced similar scrutiny. A separate analysis conducted by NewsGuard in April 2026 found that its chatbot “Le Chat” repeated Russian falsehoods in 50 percent of English-language tests and 56.6 percent of French-language tests. Those findings included misinformation relating not only to Russia but also to Iran and China.
The NewsGuard report also pointed to the influence of large-scale disinformation networks such as “Pravda” or “Portal Kombat,” a system of hundreds of websites designed to amplify pro-Russian narratives and potentially influence search engines and AI training data. Researchers suggested these networks may be contributing to the presence of biased or misleading information in AI-generated responses.
Mistral was contacted for comment but had not responded at the time of publication.
The findings come at a sensitive moment for the company, which earlier this year signed an agreement with the French government to integrate its AI systems into military and public-sector operations, including defence-related applications.
The study adds to growing concerns across Europe about the resilience of large language models to coordinated disinformation campaigns, particularly as governments and institutions increasingly rely on AI systems for information processing and decision support.
