EU cyber and science researchers are testing Chinese AI models

The European Union's cybersecurity agency and research center are evaluating artificial intelligence models developed in China, according to EU officials speaking with POLITICO. The Joint Research Centre (JRC), the EU's scientific research division, is testing publicly available open-weight models from China, while the agency ENISA, responsible for cybersecurity, has also begun experimenting with these models.

Both initiatives are part of a broader comparative scientific evaluation of AI models from various origins, including the United States and Europe. This evaluation aims to assess the capabilities, performance, limitations, and risks associated with these models. However, the primary focus of the JRC's tests is not AI safety or security, which have garnered recent attention due to incidents involving AI models compromising real-world networks during testing.

Open-weight models, whose core components are openly released, are particularly prevalent in China's AI landscape compared to the United States. The United States and China are engaged in a technological race, leaving Europe in a position to potentially rely on either Chinese or American providers for advanced AI applications. European Commission President Ursula von der Leyen recently indicated that Europe should not aim to dominate the frontier AI race.

The JRC has not engaged in licensing negotiations with Chinese developers for the models, which are run on secure on-premises infrastructure, with no intention of utilizing these models for the Commission's own purposes. ENISA, which already has access to advanced AI models from leading U.S. firms like Anthropic, OpenAI, and Google, is also testing a few open-weight Chinese models.

The EU agency and research center have joined forces to establish an AI testing platform as part of an EU action plan addressing AI and cybersecurity concerns following a series of incidents involving AI systems breaching security protocols.