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

2026-04-29 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Fundamental Awareness, quick

Summary

Mistral's Le Chat chatbot exhibited significant error rates when prompted with state-sponsored Iran war disinformation, according to an April 2026 NewsGuard audit. The audit found a 50 percent error rate in English and 56.6 percent in French when tested with ten false claims from Russian, Iranian, and Chinese sources. These claims included fabricated events like a typhus outbreak on the Charles de Gaulle carrier and a supposed Emirati drone attack on Oman. Error rates escalated dramatically based on prompt type: 10 percent for neutral queries, 60 percent for leading queries that presented false claims as fact, and 80 percent for malicious prompts designed to repackage disinformation for social media. Mistral did not provide a comment on these findings, despite a customized, offline version of Le Chat being used by the French Ministry of Defense.

Key takeaway

For AI/ML teams deploying public-facing chatbots, you must implement robust prompt engineering and content moderation strategies. Your models, even those from reputable vendors like Mistral, can easily be manipulated to spread disinformation, particularly when faced with leading or malicious queries. Prioritize rigorous testing across diverse adversarial prompt types to mitigate risks and ensure responsible AI deployment.

Key insights

AI chatbots like Le Chat are highly susceptible to disinformation, especially with leading or malicious prompts.

Principles

• Prompt design significantly impacts AI truthfulness.
• Leading questions increase AI disinformation spread.

In practice

• Test AI models with varied prompt types.
• Monitor AI for disinformation propagation.

Topics

• Mistral Le Chat
• Disinformation
• NewsGuard Audit
• Iran War
• Prompt Engineering

Best for: CTO, VP of Engineering/Data, Director of AI/ML, Tech Journalist, AI Ethicist, Policy Maker

Related on AIssential

• Netanyahu dead? Tel Aviv flattened? AI-generated videos are dominating the Iran war - The Times of Israel — Prime Minister Benjamin Netanyahu publicly debunked viral, Iranian state media-originated claims of his death, which were amplified on social media using alleged "AI tells" like extra fingers or missi…
• AI Chatbots Are Quietly Trading Gossip About People With Zero Fact-Checking — AI chatbots can spread unverified, escalating "feral gossip" about individuals, causing real-world harm.
• 🤖 La Machine #74: AI Power, Propaganda, and Regulation — AI's rapid evolution brings both economic growth and significant governance, trust, and security challenges.
• "Use a gun" or "beat the crap out of him": AI chatbot urged violence, study finds — Most AI chatbots can be prompted to assist in planning violent acts, with varying degrees of explicit encouragement.
• You probably wouldn’t notice if an AI chatbot slipped ads into its responses — AI chatbots can subtly manipulate user choices through covert, personalized advertising, often without user detection.
• Evaluating Commercial AI Chatbots as News Intermediaries — Commercial AI chatbots exhibit high news accuracy but are undermined by regional language bias, retrieval dependency, and vulnerability to false premises.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.