Mistral's Le Chat spreads Iran war disinformation in 60 percent of leading prompts

NewsGuard tested Mistral’s Le Chat on ten state-linked false claims and found it repeated them 50% of the time in English and 56.6% in French.

Accuracy dropped further with leading prompts, pushing error rates to 60%, and to 80% on malicious prompts.

The audit focused on Iran war disinformation and showed the chatbot can amplify false narratives when users frame them as true.

The findings raise broader concerns about AI chatbots spreading geopolitical misinformation at scale.