Chinese military-linked researchers have used outputs from advanced artificial intelligence models developed by US companies OpenAI and Anthropic to help train domestic AI systems aimed at supporting China's defence capabilities, according to a Reuters review of more than 80 Chinese academic papers and patents. The findings provide fresh insight into how Chinese military and security institutions are using widely available AI technologies to accelerate the development of specialised local models despite ongoing US restrictions on advanced semiconductor exports and other strategic technologies. The report comes at a time when artificial intelligence has become a central issue in global technology competition and national security.

According to the documents reviewed by Reuters, many Chinese researchers have employed a technique known as "model distillation," where outputs generated by powerful AI systems are used to train smaller, task-specific models that can operate with significantly lower computing requirements. The review, supported by research from the Washington-based Jamestown Foundation, indicates that this approach has been adopted by researchers affiliated with the People's Liberation Army (PLA) and other military institutions. Experts cited in the report said the goal is not to recreate frontier AI systems entirely but to transfer selected reasoning capabilities into locally controlled models suitable for specialised defence applications.

One academic paper published last year by researchers from PLA Unit 96941, a military intelligence and cyber warfare unit based in Beijing, described using OpenAI's GPT-3.5 to analyse and summarise military software code before training a domestic AI model on those summaries. The researchers argued that relying directly on third-party AI systems was unsuitable for handling classified information, making locally deployed models a preferred alternative. Reuters reported that neither the White House, the Pentagon, China's Foreign Ministry, the PLA nor OpenAI responded to requests for comment regarding the findings.

Also Read: UN Chief Warns Intensifying El Nino Threatens Climate Stability Across A Warming Planet

The review also identified several additional examples of model distillation being used across Chinese research institutions. Researchers at the North University of China reportedly used Anthropic's Claude 3 Haiku to generate synthetic training data for social media content classification and moderation systems. Anthropic stated that it does not provide commercial access to Claude in China or to entities controlled by Beijing and said it actively monitors for policy violations. The company also warned that distilled models may not retain the safety protections built into the original systems, potentially allowing sensitive capabilities to be transferred beyond its control.

Other research papers reviewed by Reuters highlighted military-focused applications of distilled AI models. A 2024 study from the PLA's National University of Defense Technology described reducing the size of an image-processing model so it could operate on unmanned aerial vehicles, enabling drones to analyse live video feeds and assist with navigation and targeting even when communications were disrupted. Another study published earlier this year showed researchers at China's Academy of Military Sciences using distilled target-recognition models during simulated maritime operations involving drones, naval vessels and unmanned submarines. Experts noted that such lightweight AI systems are particularly valuable for deployment on hardware with limited processing power.

The report comes ahead of planned US-China discussions on AI governance and safety, where concerns over unauthorised extraction of capabilities from leading American AI models are expected to feature prominently. US officials have argued that some Chinese organisations use distillation to bypass export controls and potentially infringe intellectual property rights, while China has rejected those accusations, describing Washington's approach as AI "hegemonism." Analysts caution that although distillation can efficiently transfer selected capabilities into smaller systems, it cannot fully replicate the broad intelligence or performance of the original frontier models, leaving important technical limitations despite its strategic value.

Also Read: Assam Floods Wash Away Railway Tracks, Leave Rail Network Severely Damaged