AI·July 31, 2026·2 min read

Investigation Finds Chinese Military Researchers Used U.S. AI Model Outputs To Train Defense Systems

Report Raises New Questions About AI Export Controls And National Security

AI neural network concept illustration.
IllustrationAI neural network concept illustration.No Spin Media AI Illustration (generated with Gemini)

An investigation by Reuters found that Chinese military researchers affiliated with institutions supporting the People's Liberation Army (PLA) have used outputs from advanced American artificial intelligence models developed by OpenAI and Anthropic to help train domestic AI systems for defense-related research. The review examined more than 80 academic papers and patent filings describing how researchers employed a technique known as model distillation, allowing smaller locally developed models to learn capabilities from larger frontier AI systems while operating independently on Chinese infrastructure.

According to the investigation, researchers used outputs generated by models including GPT-3.5 and Claude 3 Haiku to create training data, summarize complex information and improve specialized AI applications. Experts say model distillation can significantly reduce the computing power required to build capable AI systems, enabling organizations to develop models that run on domestic hardware despite U.S. restrictions on advanced semiconductor exports. The findings highlight how access to publicly available AI services may allow foreign researchers to benefit from leading American technology even when physical chip exports remain restricted.

Reuters reported that several papers linked the research to institutions with military or defense ties, while analysts at the Jamestown Foundation said the work demonstrates how frontier AI systems can indirectly contribute to military modernization through publicly accessible outputs rather than direct transfers of software or hardware. OpenAI and Anthropic have both stated that they prohibit military misuse of their services and continue monitoring for activity that violates their policies.

The report comes as the United States and China prepare for additional discussions on artificial intelligence governance, export controls and national security. Policymakers have increasingly focused on whether existing regulations adequately address access to powerful AI models through cloud-based services, particularly as model distillation techniques become more widespread across both commercial and research environments.

Community discussion
"Do you think governments should impose additional restrictions on access to advanced AI models when national security concerns are involved?"