Key Highlights
- Reuters reviewed over 80 Chinese papers and patents linked to military AI research.
- Researchers reportedly used outputs from OpenAI and Anthropic models for AI distillation.
- The technique helps build smaller AI models that run locally on military hardware.
- The findings come as AI competition between the US and China continues to intensify.
Chinese military researchers have reportedly used outputs from leading US artificial intelligence models, including those developed by OpenAI and Anthropic, to train domestic AI systems for defence applications, according to a Reuters investigation.
Reuters said it reviewed more than 80 Chinese academic papers and patents, alongside research compiled by the Washington-based Jamestown Foundation. The documents suggest researchers linked to the People’s Liberation Army (PLA) and other defence institutions relied on a technique called “model distillation” to develop specialised AI models for military use.
What is AI distillation?
Model distillation is a widely used AI training method where a smaller model learns from the outputs of a larger and more capable AI system. Instead of building a frontier AI model from scratch, developers use the responses of advanced models to train lighter versions that require significantly less computing power.
According to Reuters, Chinese military researchers used this approach to create AI systems that could operate securely within domestic military networks. These smaller models can also run on devices with limited computing resources, such as drones, surveillance systems and tactical hardware.
The report notes that distillation itself is not illegal and remains a common practice across the AI industry. However, US officials have increasingly raised concerns that some Chinese organisations may be using the technique to extract capabilities from American AI models in ways that undermine export controls and intellectual property protections.
AI research linked to military applications
Reuters identified several research papers connected to Chinese defence institutions.
One paper reportedly described researchers from PLA Unit 96941, a military intelligence and cyber warfare unit, using OpenAI’s GPT-3.5 to summarise sensitive military source code. They then trained a domestic AI model using those summaries so it could operate entirely within Chinese military infrastructure.
Another study from the PLA National University of Defense Technology explained how AI distillation helped shrink an image-processing model for deployment on unmanned aerial vehicles. The AI could analyse live video streams and assist with navigation and targeting even when communication links were unavailable.
Reuters also reported that researchers at China’s Academy of Military Sciences used distilled AI models for target recognition during simulated maritime operations involving drones, ships and unmanned submarines.
Separately, researchers at the North University of China, which has close ties to the country’s weapons industry, reportedly used Anthropic’s Claude 3 Haiku to generate synthetic training data for content moderation and social media monitoring.
Anthropic told Reuters that it does not provide commercial access to Claude in China or to Beijing-controlled entities and uses monitoring systems to detect policy violations.
AI rivalry continues to intensify
The findings arrive as artificial intelligence becomes an increasingly important area of strategic competition between the United States and China.
Washington has imposed export controls on advanced AI chips and other critical technologies to slow China’s AI development. Meanwhile, Chinese researchers have increasingly focused on lightweight AI models that can run efficiently on edge devices despite limited access to cutting-edge computing hardware.
Experts, however, caution that distilled models cannot fully match the capabilities of the frontier AI systems they learn from.
Trevor Koverko, co-founder of AI data company Sapien, told Reuters that distillation transfers selected capabilities into a smaller and cheaper model rather than replicating the full intelligence of the original system.
Neither the White House, Pentagon, China’s Foreign Ministry, the PLA nor OpenAI responded to Reuters’ requests for comment.
Source: Reuters
