Mistral AI Model Fails to Combat Russian Disinformation Effectively

A study by the Institute of the Estonian Language evaluated 60 generative AI systems for their ability to counter Russian propaganda. Mistral, a French AI company, ranked 47th, with all its models scoring below 40% in identifying Russian disinformation. In contrast, other models, including those from Anthropic and certain Chinese systems, performed better. The research highlights the growing threat of Russian propaganda, which has surged from dozens of articles per day in 2023 to nearly 10,000 in 2026. The study involved testing the models with 75 questions in multiple languages to assess their resilience against manipulation. Mistral defended its performance by stating that the study used unoptimized models and emphasized its filtering capabilities. The implications of these findings are significant for governments and organizations relying on AI for information verification.

Key Points: • Mistral ranked 47th out of 60 AI models in identifying Russian propaganda. • All Mistral models scored below 40% in recognizing malicious sources. • The rise of Russian propaganda has increased dramatically, impacting European elections.