Chinese Military Researchers Leverage U.S. AI Models for Defense Development

Chinese military researchers are continuing to strengthen the country's defense capabilities by leveraging advanced artificial intelligence (AI) models developed by leading U.S. technology companies, despite strict American restrictions on access to such technologies. By utilizing the outputs generated by AI models from companies such as OpenAI and Anthropic, Chinese researchers are reportedly developing indigenous technologies for surveillance, cyber warfare, drone operations, and strategic military missions.

The findings were revealed in an investigative report published by the international news agency Reuters.

According to information compiled by the Washington-based Jamestown Foundation and an analysis of more than 80 Chinese academic papers and patents, institutions affiliated with the People's Liberation Army (PLA) have been using advanced U.S. AI models while circumventing existing American restrictions.

Model Distillation as a Technological Shortcut

The report states that Chinese military researchers are primarily relying on a technique known as model distillation. This process uses the outputs of a large and highly capable AI system to train a smaller, specialized model that can operate locally with far less computing power.

With the United States maintaining strict export controls on advanced semiconductor chips and related technologies, Chinese defense organizations reportedly view American AI models as both a source of technical insight and a shortcut to narrowing the technological gap with the United States.

The reviewed research papers suggest that Chinese researchers have gone beyond simply obtaining accurate answers. They have also sought to absorb the underlying reasoning processes and latent strategic logic employed by Western AI models.

Sensitive Code and Military Applications

The analysis indicates that PLA Unit 96941, a Beijing-based military intelligence and cyber warfare unit, has used OpenAI's models to process sensitive military source code.

Because third-party AI models are not considered suitable for directly handling classified information, researchers reportedly used the U.S. models to generate summaries of software code before transferring the distilled knowledge to a domestically developed AI model operating within a secure military network.

In addition, researchers at North China University reportedly used Anthropic's models to generate synthetic training data for social media monitoring and content moderation systems.

Meanwhile, the Academy of Military Sciences is said to have employed the technology to improve target identification capabilities for drones, naval vessels, and unmanned submarines during maritime exercises.

Rising Geopolitical Tensions

The revelations have intensified geopolitical tensions ahead of upcoming discussions between the United States and China on AI governance and security.

U.S. officials have alleged that such unauthorized use violates intellectual property rights and circumvents American export controls. Beijing, however, has rejected those accusations, countering that Washington is attempting to establish unilateral dominance in the field of artificial intelligence.

//DBTech/BMT/OR//