Chinese military-linked researchers have utilized outputs from leading American artificial intelligence models—including those developed by OpenAI and Anthropic—to build specialized domestic systems for defense, surveillance, and cyberwarfare. According to a review of over 80 academic papers and patents compiled by the Jamestown Foundation, Chinese security institutions are deploying a technique called “model distillation”. By training smaller, localized software on data generated by frontier U.S. systems, researchers can replicate complex reasoning capabilities while bypassing Washington’s stringent export controls on high-end hardware.
The distilled systems are being deployed across several tactical military domains. Researchers from People’s Liberation Army units have used Western model outputs to develop image-processing algorithms for unmanned aerial vehicles, allowing drones to analyze live video and execute targeting decisions offline. Other projects utilized model outputs to analyze sensitive source code and generate synthetic data for content moderation and social media monitoring. These lightweight models enable Chinese forces to run specialized AI applications locally on edge devices with limited computing power.
The findings come amid escalating diplomatic tensions between Washington and Beijing regarding international AI safety and intellectual property rights. U.S. officials contend that unauthorized distillation undercuts technology controls, while Chinese representatives reject the accusations and accuse Washington of pursuing technological hegemonism. While industry analysts caution that distilled models remain dependent on foreign foundational systems, the practice underscores how open research outputs continue to cross geopolitical barriers.
