Chinese military researchers have repeatedly used outputs from leading American artificial intelligence models to train domestic AI systems, according to a review of studies compiled by the Washington-based Jamestown Foundation and independently verified by Reuters. The technique, known as model distillation, involves a smaller “student” model learning from a larger “teacher” model, enabling the transfer of selected capabilities into cheaper, locally controlled systems. One paper from PLA Unit 96941, a military intelligence and cyber warfare unit, described using OpenAI’s GPT-3.5 to analyse military software code before transferring that knowledge into a domestically developed model capable of operating entirely within Chinese military networks. Other research outlined the use of Anthropic’s Claude AI to generate training data for social media monitoring and content moderation systems.

Viewed from Washington, the practice raises fresh concerns over national security and the effectiveness of export controls. US officials have accused some Chinese organisations of using distillation to extract capabilities from proprietary American AI systems, potentially weakening intellectual property protections. Anthropic stated it does not provide commercial access to Claude in China and warned that distilled models could lose important safety protections while retaining valuable capabilities. Beijing has rejected the allegations, accusing Washington of attempting to maintain AI dominance. Sunny Cheung, a fellow at the Jamestown Foundation who analysed more than 60 papers, said the research points to a coordinated effort to narrow China’s AI gap with the United States, noting that “these papers show Chinese military-linked researchers are trying to transfer that expensive, proprietary reasoning from Western models into smaller systems they can control and deploy locally.”

The distillation push coincides with a direct order from President Xi Jinping to accelerate the military application of artificial intelligence. During a Politburo group study session reported by state news agency Xinhua, Xi called on the People’s Liberation Army to “strengthen the military application of unmanned intelligent technologies, deepen the development and application of network information systems, and gradually build an intelligent military system.” He emphasised that the fundamental standard for evaluating modernisation is combat effectiveness and reiterated that the PLA must meet its 2027 centenary goal of achieving a major breakthrough in modernised combat capability. Xi also demanded deeper anti-corruption efforts and tighter supervision of major defence projects.

Factual implications of the research are concrete. Additional studies reviewed by Reuters described Chinese military researchers using distillation to develop AI models capable of running on drones, autonomous ships and other battlefield systems with limited computing power. A paper from the PLA’s National University of Defense Technology outlined shrinking an image-recognition model so it could process live video aboard unmanned aerial vehicles without external communications. Researchers at China’s Academy of Military Sciences also described deploying distilled AI models during simulated maritime operations involving drones, warships and unmanned submarines. The technique is widely used across the AI industry, but the controversy centres on unauthorised extraction of capabilities from closed, proprietary models accessed through interfaces, rather than on distillation itself.

The dossier remains active. The Jamestown Foundation and Reuters identified dozens of military-linked examples, and Xi’s directive signals sustained institutional pressure to integrate AI into the force structure ahead of the 2027 centenary. No specific regulatory or diplomatic next steps have been announced by Washington or Beijing, but the findings are likely to fuel further debate over the adequacy of current export controls and the boundaries of legitimate AI research.