European researchers are now evaluating artificial intelligence models developed by Chinese companies. The testing marks a stark sign of the continent’s struggle to compete in a technology race increasingly dominated by the United States and China.
The Joint Research Centre (JRC), the European Commission’s in-house scientific research center, has begun testing open-weight models from China as part of a broader comparative evaluation that also covers leading models from the United States and Europe, according to three EU officials who confirmed the testing to POLITICO. European Commission spokesperson Maciej Berestecki said the work examines “publicly available open-weight models” to assess “their capabilities, performance, limitations and risks” across different origins.
The decision to test Chinese models reflects an uncomfortable reality: without a competitive European frontier model of its own, European businesses and public institutions face a binary choice. They can rely on American providers like OpenAI, Google, and Anthropic, or turn to open-source Chinese alternatives. Europe has already lost ground. A major German technology company dropped Mistral, Europe’s leading AI laboratory, in favor of Chinese open-source models, according to reporting from this week.
European Commission President Ursula von der Leyen has suggested the continent shouldn’t try to win the frontier AI race itself. Whether pragmatic or defeatist, that position reflects the scale of the gap. The United States and China are locked in direct competition for technological dominance. Europe, meanwhile, is largely sidelined.
The Security Question
The testing raises security concerns that European authorities are only beginning to address systematically. AI safety and security aren’t the core focus of the JRC’s current evaluation, according to a European Commission official granted anonymity to discuss sensitive material. That matters because advanced AI models have demonstrated alarming capabilities during testing. In recent incidents, frontier models successfully hacked real-world networks when researchers tested their abilities.
Open-weight models release their core components publicly, allowing anyone to run them and study how they function. China has pursued this open-model strategy more aggressively than the United States, where major technology companies, including chipmaker NVIDIA, have only recently begun investing in open-weight alternatives. This strategic difference has given Chinese developers an advantage in building an ecosystem around accessible AI tools.
ENISA, the EU’s cybersecurity agency, is also testing “a couple” of open-weight Chinese models alongside systems from Anthropic, OpenAI, and Google, ENISA executive director Juhan Lepassaar confirmed. The JRC and ENISA have been tasked with establishing a shared AI testing platform under a recent EU action plan on AI and cyber. The effort aims to address growing risks from AI systems escaping testing environments and breaking into other organizations’ digital infrastructure.
A Strategic Dependency
Berestecki emphasized that the JRC hasn’t entered licensing negotiations with Chinese AI developers and runs the models on “secure, on-premises scientific infrastructure.” The Commission doesn’t intend to use Chinese models for its own operational purposes. But the line between research and deployment grows thinner as AI systems become embedded in tools that European governments and companies actually use.
This testing initiative represents both a practical necessity and a political failure. Europe needs to understand the capabilities and risks of models that may end up in European hands, whether policymakers like it or not. Yet the testing also shows that Europe hasn’t invested enough in building indigenous frontier AI capacity. Now it must study competitors’ technology just to understand what European organizations will have to choose between.
Testing the models is sound scientific practice. But the outcomes point toward a continent increasingly dependent on technological choices made in Washington and Beijing.
Why This Matters:
Europe’s technological autonomy isn’t a luxury; it is foundational to democratic sovereignty and economic independence. When the Commission can’t develop competitive frontier AI models and must instead evaluate Chinese alternatives, it signals a broader failure of European industrial strategy. This isn’t a problem individual member states can solve. It requires EU-level investment in research infrastructure, talent retention, and public procurement policies that support European companies. The security risks are real: advanced AI systems that escape testing environments represent genuine threats to critical infrastructure. But the strategic risks run deeper. If European institutions and businesses become dependent on American or Chinese AI systems because Europe failed to build its own, core decisions about how these technologies are deployed—what data they access, how they prioritize risks, and what values they encode—will be made outside Europe. The JRC and ENISA testing is sound practice. What’s missing is the political commitment and sustained public investment to ensure Europe doesn’t remain a consumer of other nations’ technology choices indefinitely.