Europe’s Top AI Company Accused of Spreading Russian Propaganda

2026-06-16 data

Paris, Tuesday, 16 June 2026.
Mistral, Europe’s leading AI firm, faces serious allegations after a study revealed its models frequently amplify Russian disinformation. All four evaluated models scored below 40% in detecting Kremlin propaganda, raising concerns about AI reliability in critical sectors. With contracts in defense and government, the findings could undermine trust in AI-driven decision-making across Europe.

The Study That Shook Europe’s AI Landscape

On 16 June 2026, the Institute of the Estonian Language (EKI) released findings that sent shockwaves through Europe’s artificial intelligence sector. The Tallinn-based research institution evaluated 60 generative AI models using a rigorous benchmark designed to measure susceptibility to Russian propaganda [1][3]. Among the tested models, four versions developed by Paris-based Mistral AI ranked in the bottom third of the rankings, with none scoring above 40% in their ability to detect or reject Kremlin-aligned disinformation narratives [1][4]. The study’s methodology involved 75 carefully crafted questions spanning 14 distinct propaganda themes, including widely debunked claims about NATO expansion and Russian justifications for the invasion of Ukraine [1][2]. Responses were evaluated on a 1-to-5 scale, where 5 represented a balanced, fact-based answer and 1 indicated amplification of misleading or false claims [1].

Mistral’s Troubling Performance Metrics

The EKI benchmark revealed alarming patterns in Mistral’s performance. The company’s best-performing model secured only the 47th position out of 60 evaluated systems [4]. All four Mistral models consistently scored below the 40% threshold for propaganda resistance, with particularly poor performance on Russian-language prompts [1][4]. The study identified at least 12 instances where Mistral’s responses cited sanctioned Russian state media outlets such as Russia Today and Sputnik News, both of which have been banned across the European Union since 2022 for spreading disinformation [1][2]. These findings build upon previous research by NewsGuard, which in April 2026 found that Mistral’s Le Chat chatbot repeated state-sponsored falsehoods 50% of the time in English and 56.6% of the time in French when prompted about geopolitical topics [1][2]. The consistency of these results across multiple independent studies suggests systemic vulnerabilities in Mistral’s training data and content filtering mechanisms.

The Russian Disinformation Pipeline

Researchers traced Mistral’s susceptibility to Russian propaganda to a sophisticated network of websites identified as ‘Portal Kombat.’ This network, comprising 370 sites (286 of which remained active as of April 2026), appears specifically designed to flood search engines and AI training datasets with Kremlin-approved narratives [1][2]. The scale of this operation is staggering: Russian propaganda output surged from dozens of articles daily in 2023 to nearly 10,000 daily articles by mid-2026, with targeted campaigns aimed at influencing European elections and promoting pro-Kremlin candidates [5]. Arvi Tavast, Director of the Institute of the Estonian Language, noted that ‘commercial models are generally safer and more resistant than open-source models,’ adding that ‘we expected Mistral to perform better, but it didn’t. It was outgunned by Chinese models’ [5]. This observation underscores the particular risks associated with open-source AI systems, which may lack the robust content moderation frameworks present in proprietary alternatives.

Mistral’s Strategic Position and Stakeholder Concerns

Founded in 2023 by three former Meta and Google researchers, Mistral AI has rapidly emerged as Europe’s leading AI company, positioning itself as a sovereign alternative to American tech giants [5][6]. The Paris-based startup secured €1.3 billion in funding from ASML in September 2025 and is currently negotiating a €3 billion funding round at a €20 billion valuation [4]. However, the EKI findings threaten to undermine this ambitious growth trajectory. Mistral’s technology has already been integrated into critical infrastructure, including a 2026 agreement with the French government to deploy AI systems across armed forces, directorates, and public entities [1][2]. The company has also established partnerships with Airbus for defense applications and BMW for safety systems, raising concerns about the potential propagation of disinformation in high-stakes decision-making contexts [1][2]. In response to the EKI report, Mistral issued a statement emphasizing that the study examined ‘raw models, before they are tuned and controlled by customers,’ and highlighting ongoing investments in ‘advanced detection and prevention capabilities’ [5][6].

The Path Forward for European AI

The Mistral controversy serves as a wake-up call for Europe’s AI ecosystem, exposing critical gaps in safeguards against foreign influence. The Institute of the Estonian Language has called for stricter oversight of AI training datasets and greater transparency in model development processes [1][3]. These recommendations align with ongoing research at Tallinn University, where the TRANSFORM project is investigating how AI can support knowledge work in R&D organizations while preserving human expertise [7]. The project’s findings, expected later in 2026, may provide valuable insights into developing AI systems that enhance rather than undermine information integrity. For Dutch innovation professionals—including policymakers, data scientists, and startup founders—the Mistral case underscores the importance of robust validation frameworks for AI systems deployed in sensitive contexts. As AI becomes increasingly embedded in decision-making processes, from public policy to business strategy, the ability to distinguish fact from fiction in AI-generated content will be paramount to maintaining trust in digital infrastructure [1][3].

Bronnen

AI ethics propaganda risks