Europe's leading AI company Mistral struggles to detect Russian disinformation, according to Estonian researchers. The analysis ranked all four Mistral models below 40% effectiveness in identifying malicious Russian propaganda, with the top system placing 47th out of 60 generative AI models tested.

Commercial competitors like Anthropic's Claude and Chinese systems outperformed the Mistral models, raising concerns about open-source AI's vulnerability to state-sponsored manipulation. The institute tested models using 75 questions across three languages, focusing on 14 Russian propaganda themes including false claims about Ukraine and NATO expansion.

Mistral defended its performance, noting the study examined raw models before customer tuning and that its Vibe Work function includes filtering layers. The company highlighted its €4bn investment in Nvidia-powered data centers across France and Sweden as part of expanding European AI infrastructure.

Valued at nearly €12bn, Mistral aims to reach $1bn annual recurring revenue by end-2026. Backed by ASML with €1.3bn in funding, the company represents Europe's biggest challenger to US and Chinese AI dominance despite current security shortcomings.