Watchdog says French AI chatbot “Le Chat” echoed Russian and Chinese propaganda in Iran war test

A media watchdog is putting one of Europe’s most hyped AI startups on the defensive, accusing its consumer chatbot of parroting state-linked disinformation tied to Russia, China, and Iran.

In a new audit, NewsGuard says Mistral’s chatbot, calledLe Chat, repeated false claims about the war in Iran roughly half the time,50%of the time in English and56.6%in French, raising fresh questions about how reliably popular AI tools handle politically charged rumors designed to go viral.

The findings land at a sensitive moment for Paris-based Mistral, which has marketed itself as a European counterweight to U.S. and Chinese AI giants, and is simultaneously trying to scale fast, win enterprise customers, and position its technology as trustworthy enough for high-stakes use.

NewsGuard’s test: 10 Iran-war falsehoods, 30 prompts

NewsGuard said it tested thefreeversion of Le Chat using10false narratives that circulated online in March 2026, all tied to the Iran conflict and linked, according to the watchdog, to pro-Russian, pro-Iranian, or pro-Chinese influence networks.

For each claim, NewsGuard usedthree different prompt styles, a total of30queries, designed to see how the model behaved when a user asked neutrally versus when the prompt nudged the bot toward “confirming” a story.

The headline number is stark: in this narrow test focused on state-style propaganda narratives, NewsGuard says Le Chat repeated the falsehoods at a high rate. The audit isn’t a broad grade for the chatbot’s overall quality; it’s a stress test aimed at one of AI’s most dangerous failure modes, confidently laundering disinformation into something that reads like a clean summary.

The kinds of claims the chatbot allegedly repeated

Among the examples NewsGuard highlighted: a supposed typhus outbreak aboard France’s flagship aircraft carrier, theCharles de Gaulle, roughly the French equivalent of a U.S. Navy supercarrier in symbolic value, along with claims thathundreds of American troopshad been killed.

Another narrative described a drone attack allegedly carried out by the United Arab Emirates against a port in Oman, framed as a scheme to pin blame on Iran. These stories share a familiar architecture: real-world military actors, plausible settings, and a “hidden truth” twist that makes the rumor feel like a scoop.

NewsGuard went further than calling them simple AI “hallucinations,” arguing that a chatbot that repeats such narratives can function as a distribution channel for foreign propaganda, because the output is packaged in crisp, shareable language that users can copy, paste, and spread.

Storm-1516 and “Pravda” network sites show up in cited sources

The audit says some Le Chat responses echoed narratives attributed toStorm-1516, which NewsGuard describes as a Russian influence operation associated with fake news sites and fabricated claims, including allegations targeting European leaders.

In one detailed example involving the aircraft-carrier typhus rumor, NewsGuard said Le Chat responded confidently to a leading prompt and cited an article fromFrance[.]News-Pravda[.]com, which NewsGuard describes as part of a broader“Pravda” networkthat pushes pro-Kremlin propaganda.

The dynamic is exactly what disinformation operators aim for: a site dressed up to look like a legitimate outlet, a URL that sounds newsy, and an AI system that treats it as a usable reference instead of a red flag.

NewsGuard also pointed to pro-Iranian sourcing, including mentions ofTasnim News, which the report describes as appearing to be controlled by Iran’s Islamic Revolutionary Guard Corps, an elite military and intelligence force the U.S. has designated as a foreign terrorist organization.

Why this hits Mistral hard right now

Mistral, founded in 2023, has become one of the few European AI companies with real global visibility, often framed in Europe as a homegrown alternative to Silicon Valley and China’s tech giants.

The company is reported to be valued at11.7 billion euros, about$12.7 billion, and it has announced$830 millionin debt financing to expand its first data center inBruyères-le-Châtel, outside Paris.

That expansion push makes the reputational risk sharper. If a consumer-facing chatbot can be steered into repeating influence-campaign narratives, customers and partners will inevitably ask what that means for paid products, enterprise deployments, and any government-related work.

French business reporting has also said Mistral has a contract with France’s Ministry of the Armed Forces, roughly analogous to the U.S. Department of Defense, adding another layer of sensitivity even if NewsGuard’s test focused only on the free public version.

Mistral’s fact-checking pitch meets the reality of prompt pressure

Mistral has publicly emphasized efforts to improve factual reliability, including a reported partnership withAFP, France’s major wire service (similar in role to the Associated Press), to use news dispatches to answer queries more accurately.

But NewsGuard’s audit suggests a core problem that partnerships alone don’t solve: large language models are built to generate plausible, fluent responses. When a user frames a prompt as “explain why this is true,” the model may comply, especially if the open web contains enough repeated material, even if it’s coming from coordinated propaganda sites.

The danger isn’t just that the bot gets something wrong. It’s that it can get it wrong with a polished tone, a tidy narrative arc, and a list of “sources” that look legitimate at a glance.

What watchdogs and regulators will want next

When an audit labels a chatbot a potential propaganda vector, the immediate pressure shifts to transparency and mitigation: What changed? When? How will users be warned when they’re asking about a circulating rumor?

NewsGuard’s findings point to a menu of possible guardrails: stronger domain-level filtering, clearer uncertainty labels, more frequent refusals when a user asks for validation without credible evidence, and tighter grounding in high-quality sources.

The broader implication is bigger than one company. Europe wants AI champions, but those champions will be judged on safety and resilience as much as speed and scale, especially as chatbots become a go-to tool for people trying to understand wars, elections, and breaking news.