Chinese military researchers have used outputs from leading American artificial intelligence models — built by OpenAI and Anthropic — to train domestic AI systems intended to advance China's defense capabilities, according to a Reuters exclusive picked up this week by The Japan Times, The Hindu, Defense News and outlets around the world.
Reuters reported reviewing more than 80 Chinese papers and patents showing military-linked institutions drawing on the technique. The documents describe applications including surveillance, cyber warfare and battlefield systems.
The method at the center of the reporting is what the documents call "model distillation." In plain terms, it means using the answers a powerful AI system produces as training material for a smaller, cheaper, more specialized model. You don't need the original model's internal machinery — just enough of its output to teach a new system to imitate it. As one summary of the reporting put it, when high-end hardware is hard to obtain, workarounds become valuable.
That is the crux of why the story landed hard. Washington has spent years restricting China's access to advanced American chips and AI technology. Reporting aggregated by MSN and Business Insider Africa frames the findings as evidence those restrictions failed to keep American AI capabilities out of Chinese military hands, even as Beijing expands arms sales across Africa.
The timing sharpens the point. Bloomberg reported separately that Xi Jinping has urged China's military to step up its use of unmanned and AI technologies.
Neither the companies' responses nor any confirmation from Beijing appears in the source items available here, and the reporting is described throughout as based on documents Reuters reviewed.
It matters because it suggests export controls aimed at hardware may not stop the thing they were meant to stop: if a model's outputs can be copied and repurposed, capability leaks through the front door.