A few weeks A few months ago, the French company Mistral released a first version of its new artificial intelligence model, nicknamed Le Chonk. At the time, the company said it was the most advanced open weighting offering outside of China. After further improvements, Mistral now claims that Le Chonk is the world’s “best” open model for automated coding and cybersecurity.
According to Mistral, this progress was made possible by feeding the model increasingly difficult cybersecurity and coding issues, allowing the trillion-parameter Le Chonk to match or surpass models from leading Chinese labs such as DeepSeek and Z.ai.
“This is a pivotal moment in the race for global leadership in AI,” Pierre Stock, vice president of science at Mistral, told reporters at a press briefing. “It provides businesses and governments with cutting-edge capabilities to defend against an increasing level of cyber threat.”
The finalized version of Le Chonk is now freely available to everyone. After a series of AI-related cybersecurity incidents, Mistral says it is important for the West to have its own advanced and open models. “You have no guarantee that (Chinese models) won’t behave maliciously. They can be threat vectors,” Stock said.
The most high-profile episode occurred when one of OpenAI’s models broke loose from a testing sandbox and hacked several companies over the summer. This has sparked panic that AI could escape human oversight and be exploited by malicious actors to launch sophisticated cyberattacks. (One of the affected companies, Hugging Face, used a Chinese open-weighted model to evaluate and defend against the attack.)
Since then, AI has also uncovered security flaws in ubiquitous software. In October, a series of websites were taken offline after an AI tool discovered serious flaws in a fundamental part of web technology.
Mistral presents its open weight models, like Le Chonk, as part of the solution. If AI is to become a central part of both cyberattack and defense, businesses and governments will need to be equipped with models that are freely available and cannot be disabled remotely, unlike proprietary models like OpenAI, Anthropic, and Google.
“AI is going to be like electricity. You don’t want to live in a world where your neighbor can cut off the power any day,” Stock said. “Once you open the weights, they’re all over the Internet. It’s impossible to cut off access to them.”
The trade-off, however, is that freely available templates can also be downloaded and customized by malicious actors.
“This is clearly a dual-use technology,” Rafe Pilling, senior director of threat intelligence at security firm Sophos, told WIRED. “Anything that can be downloaded and distributed unhindered can be misused and used for productive purposes. »
According to Pilling, while it is possible for a Chinese model with Internet access to siphon data, all advanced models are capable of becoming malicious without careful monitoring, regardless of their country of origin.
“The technical realities and the narrative – which often has a nationalist bent – are two different things,” he says. “It’s about putting the appropriate monitoring and controls in place to make sure that even if the model was trying to be deceptive or harmful, you might have a way of detecting it. »
Gn bussni
