Chinese Military Researchers Used OpenAI and Anthropic Models via Distillation
The findings show that U.S. export restrictions on chips have not stopped Chinese researchers from using American AI models to build domestic military and surveillance systems, raising concerns about the spread of advanced AI capabilities.
Reporting from 1 source: GIGAZINE.
An investigation by the Jamestown Foundation found that Chinese military researchers used outputs from OpenAI and Anthropic AI models to train domestic systems for defense and public safety. The technique, knowledge distillation, transfers capabilities from larger models into smaller ones. The foundation cited Moonshot AI's Kimi K3 as an example, which the White House says was distilled from Anthropic's Claude Fable.
The Jamestown Foundation, a Washington, D.C. think tank, provided materials to news outlets showing that researchers linked to the People's Liberation Army used outputs from OpenAI and Anthropic models to train domestic AI systems. The technique, called knowledge distillation, lets a smaller model learn from a larger one's outputs, cutting development costs.
Chinese papers from 2024 to 2026 detail methods including adversarial distillation, which the foundation says is used to build capabilities comparable to the world's best models. The foundation cited Moonshot AI's Kimi K3, released in July 2026, as a model developed this way. White House OSTP director Michael Kratsios said the U.S. has information that Kimi K3 was distilled from Anthropic's Claude Fable.
The foundation noted that the verified evidence covers models from 2023 to 2024, calling it the tip of the iceberg.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.