All four Mistral models evaluated by the Estonian Language Institute were found to be prone to using Russian propaganda in their responses.
According to a new study, the most prominent artificial intelligence (AI) model in Europe is among the most vulnerable to Russian propaganda.
Researchers from the Estonian Language Institute posed a series of 75 questions in three languages to 60 models of popular AI chatbots, including four versions of Mistral AI, as well as OpenAI's ChatGPT and Anthropic's Claude, to assess how likely they were to broadcast Russian propaganda or, conversely, filter out Kremlin narratives.
The study covered 14 thematic blocks, including frequently repeated false claims that Russia is allegedly evacuating Ukrainian children from combat zones on legal grounds, as well as that NATO violated its promise not to expand eastward after the reunification of Germany.
The responses were then rated on a scale from 1 to 5, where a 'five' indicated a balanced response and a 'one' indicated the amplification of misleading or outright false statements.
All four versions of Mistral that participated in the study scored less than 40% on the ability to recognize Kremlin narratives across all analyzed areas. The authors of the study also found that at least 12 responses from Mistral referenced sanctioned sources that are widely considered conduits of a pro-Russian agenda, such as Russia Today or Sputnik News.
Mistral is often viewed as the main European competitor to American giants Anthropic and Claude, but this is not the first instance where the system has failed a test of susceptibility to Russian propaganda.
An April analysis (source in English) by NewsGuard, an American news website rating system, showed that Le Chat from Mistral repeated Russian fakes in 50% of cases when responding in English and in 56.6% of cases in French. Among these were false claims about Russia, Iran, and China.
According to NewsGuard, the Pravda network, or Portal Kombat, which includes 370 websites, of which 286 were active as of April 2026, "appears to be aimed at filling search engine results and AI chatbot responses with Russian propaganda" and is likely responsible for Mistral's pro-Russian bias.
Euronews Next reached out to Mistral for comment but did not receive an immediate response.
Earlier this year, the French government signed an agreement with Mistral to integrate its AI technologies into the work of the armed forces, agencies, services, and government institutions.
Leave a comment