Mistral says Le Chonk has moved ahead of Chinese open models

Why Mistral is making the case

Mistral credits Le Chonk’s improvements to training on increasingly difficult cybersecurity and programming tasks. Pierre Stock, the company’s vice president of research, called the progress an important moment in the global race for AI leadership. He said the model gives companies and government agencies advanced capabilities to defend against growing cyber threats.

After several incidents involving AI and cybersecurity, Mistral argues that Western countries need advanced open models of their own. Stock warned that there is no guarantee Chinese models will behave safely and said they could become a source of threats.

That argument is tied to a recent sequence of incidents:

Access cuts both ways

Mistral presents open models such as Le Chonk as part of the answer. If AI becomes a tool for both cyberattacks and cyberdefense, the company argues, businesses and public agencies need models they can use freely and that cannot be switched off remotely. Owners of closed models, including those from OpenAI, Anthropic and Google, can withdraw access.

Stock compared AI to electricity: he said nobody would want to live in a world where a neighbor could cut off their power at any moment. Once model weights are published, they spread across the internet and access cannot be shut down.

The same permanence gives attackers access, too. Sophos threat intelligence senior director Ray Pilling called the technology dual-use: a freely distributed model can be used for good or harm.

Pilling also cautioned against treating national origin as a reliable measure of risk. A Chinese model with internet access could, in theory, gather data, he said, but any advanced model can behave unpredictably without careful oversight, regardless of where it comes from. Technical properties and nationalistic interpretations of a model’s origin are separate issues. Proper monitoring and controls, he said, can help detect deceptive or malicious behavior.

The claim that matters more

I think Mistral’s strongest point is about access, not national competition: institutions may want AI they cannot lose at a provider’s discretion. But the incidents in the story make the trade-off plain. A model that cannot be remotely withdrawn is also harder to contain once its weights are out.

What I’d want to know is how Le Chonk’s reported gains were measured, and what oversight is needed to make those capabilities safe in practice. The announcement makes a case for keeping open models available; it does not establish that availability alone makes them dependable.