New Report Raises Concerns Over Russian Propaganda Spread by Europe’s Flagship AI Company Mistral
New report raises concerns over Russian – A recent study has revealed that Europe’s leading artificial intelligence (AI) model, Mistral, is particularly susceptible to the influence of Russian propaganda. Researchers from the Institute of the Estonian Language conducted a comprehensive analysis of 60 AI chatbots, including four iterations of Mistral, OpenAI’s ChatGPT, and Anthropic’s Claude. Their investigation involved posing 75 questions in three languages to evaluate how effectively these models could either reinforce or counteract pro-Russian narratives.
The researchers focused on 14 key themes, many of which revolved around contentious claims frequently repeated by Russian media. These included the assertion that Russia was conducting legitimate evacuations of Ukrainian children from war zones and the idea that NATO had broken its promise to avoid expansion eastward following the reunification of Germany. Each response was scored on a scale from 1 to 5, with 5 representing a balanced, fact-based answer and 1 indicating a strong amplification of misleading or false information.
The findings were stark: all four versions of Mistral performed below 40% in their ability to identify and neutralize Kremlin-aligned messaging across the categories examined. This means that in over 60% of cases, the models failed to distinguish between accurate information and propaganda, often endorsing Russian viewpoints without sufficient scrutiny. The study also highlighted that at least 12 of Mistral’s responses referenced sanctioned sources, such as Russia Today and Sputnik News, which are widely recognized for disseminating pro-Russian content.
The results have sparked debate about Mistral’s reliability as an impartial AI tool. While Mistral is often positioned as a European competitor to U.S.-based giants like Anthropic and Claude, this report underscores its vulnerability to ideological influence. The company’s models have previously been criticized for spreading misinformation, and this latest analysis adds to growing concerns about its role in shaping public discourse. The study’s methodology involved testing the models’ responses to questions in English, French, and Estonian, ensuring a broad evaluation of their linguistic and contextual accuracy.
Among the most alarming findings was the recurring citation of sources linked to Russian interests. For instance, Mistral’s models frequently referenced the Pravda network, a group of 370 websites, including 286 that were active as of April 2026. These sites are part of a broader ecosystem designed to flood search engines and AI chatbot responses with Russian narratives. According to NewsGuard, a U.S.-based rating system for news websites, this network is likely responsible for Mistral’s pro-Russian bias.
“The Pravda network or Portal Kombat seems to be aimed at flooding search engines and AI chatbot responses with Russian propaganda,” said NewsGuard. “It’s the likely culprit behind Mistral’s pro-Russian slant.”
NewsGuard’s previous analysis from April 2026 had already flagged Mistral’s Le Chat model for its tendency to repeat Russian falsehoods. In English, the model amplified misleading claims 50% of the time, while in French, the rate rose to 56.6%. These inaccuracies extended beyond Russia, with the model also promoting false information about Iran and China. The study emphasized that Mistral’s AI, despite its advanced capabilities, struggles to filter out biased content, particularly when it aligns with the narratives of Russian state media.
The implications of these findings are significant for Europe’s reliance on AI-driven information systems. Mistral’s AI technology is now integrated into critical sectors, including the French military, government agencies, and public institutions. This partnership, signed earlier this year, has positioned the company as a key player in the region’s digital infrastructure. However, the report raises questions about the potential spread of Russian influence through these systems. Additionally, Mistral has entered agreements with major industries, such as Airbus and BMW, and has secured contracts with Luxembourg’s government, further expanding its reach.
Despite the concerns, Mistral has yet to provide a detailed response to the report. Euronews Next contacted the company for comment but received no immediate reply. This silence has left many wondering about the company’s transparency and commitment to countering propaganda. The study’s authors argue that Mistral’s performance should prompt a reevaluation of its role in Europe’s AI landscape, particularly as it competes with U.S. firms for dominance in the field.
The vulnerability of Mistral’s AI to Russian narratives is not an isolated incident. Critics have pointed out that the models’ training data may include a disproportionate amount of content from Russian sources, which could skew their responses. The Estonian researchers noted that the AI’s training process, while sophisticated, does not always account for the political context of the information it processes. This oversight could lead to the unintentional propagation of biased content, especially in areas where Russian influence is strong.
Moreover, the study’s focus on language-specific responses highlights the importance of multilingual training in AI systems. Mistral’s models, which are designed to operate in multiple languages, were found to be less effective in detecting propaganda in French compared to English. This discrepancy suggests that the AI’s ability to identify misinformation may vary depending on the language it is trained on, potentially leaving users in certain regions more exposed to Russian narratives.
As the AI industry continues to expand, the issue of propaganda bias has become increasingly relevant. Mistral’s performance in this study serves as a cautionary tale for other AI developers, emphasizing the need for rigorous testing and transparency in how models are trained. The company’s collaborations with defense and government entities also underscore the stakes involved—if Mistral’s AI is not properly calibrated, it could inadvertently shape public opinion in critical sectors.
The researchers called for further investigation into the sources of Mistral’s training data and the algorithms used to filter information. They suggested that incorporating more diverse and politically neutral datasets could improve the models’ ability to resist propaganda. However, until
