EU's Cyber Agency and Research Centre Test Chinese AI Models
European Union flags outside the Berlaymont building in Brussels Credit: © Travel_Motion / Getty Images via Canva.com
The European Union's cyber agency, ENISA, and the Commission’s science service, the Joint Research Centre (JRC), have initiated testing of open AI models from China, as reported by Politico. These open-weight models are fully released, allowing anyone to download, run, and study them.
French Mistral, a new Large 4 model, is pitched as the strongest open-weight system built outside China this week.
ENISA and the JRC are evaluating Chinese models alongside leading American and European systems, with a focus on their capabilities, limitations, and potential risks. The JRC runs these models on its secure computers without seeking special access or licenses from Chinese developers.
According to ENISA's executive director, Juhan Lepassaar, the agency is testing a couple of Chinese open-weight models. They already have access to advanced models from Anthropic, OpenAI, and Google, capable of finding software flaws at scale.
Both organizations are also developing an AI testing platform together as part of the Commission's AI and cybersecurity plan from July. This platform will allow sectors like energy, finance, and health to test AI tools on simulated networks without risking real systems.
While safety and security are not the primary focus of these tests, they come at a time when Europe is considering its reliance on foreign AI models. Without a top-tier European model, Europe might have to choose between American and Chinese providers.
This week, Politico also reported that a large German tech firm had switched from Mistral to Chinese open models.