Le Chat repeats Iran war disinformation in 60% of leading prompts

What changed

A NewsGuard audit found Mistral's Le Chat repeated disinformation about the Iran conflict in 60% of tested prompts. The repeat rate ranged from 10% on neutral queries up to 80% on prompts intentionally designed to elicit malicious responses. The audit covered a spectrum of geopolitical content typical of news-adjacent product workloads.

Why it matters

If you build apps or integrate APIs around Mistral, this is a hard signal that the base model lacks grounding on volatile current events. Probabilistic generation without retrieval grounding will surface state-sponsored narratives and confidently wrong claims. Any product that touches news, geopolitics, or public-facing content is exposed to reputational and legal risk if you ship on Le Chat without a verification layer.

What to watch for

Monitor Mistral for a public response, model version bump, or alignment patch. Until then, route sensitive topics through a RAG pipeline against trusted journalism sources, or fall back to a provider with tighter grounding. If you maintain a Le Chat integration, audit your prompts for any current-events surface area and add a fact-check gate before output reaches users.