ENISA conducts secret tests on Chinese AI models to protect critical infrastructure

The European Union Agency for Cybersecurity (ENISA) is conducting covert stress tests on open-weight artificial intelligence models originating from China. To mitigate risks of data leaks and espionage, ENISA is testing these models on air-gapped, isolated networks that are disconnected from the internet. This approach aims to bypass potential data exposure risks associated with China’s 2017 National Intelligence Law, which requires Chinese companies to cooperate with state intelligence requests.

ENISA Executive Director Juhan Lepassaar confirmed the testing of these models within secure internal infrastructures, though specific models and methodologies remain classified. The evaluations seek to identify security vulnerabilities embedded within model weights and to prepare for new legal oversight mechanisms to protect Europe’s critical infrastructure.

Simultaneously, the broader cybersecurity landscape is shifting as attackers increasingly use AI to automate the exploitation of vulnerabilities. Rather than focusing on single flaws, attackers are utilizing AI to analyze technical data and link seemingly low-risk issues—such as API vulnerabilities and cloud misconfigurations—into complex, high-speed attack chains. This evolution necessitates a shift in defense strategies from simple patch management to comprehensive end-to-end attack path analysis.