China employs American AI in developing its military capabilities despite Washington's restrictions

A Reuters review reveals that Chinese researchers linked to the military used outputs from American AI models via "model distillation" to train local systems for military purposes, bypassing U.S. restrictions on exporting advanced chips.

A Reuters review of more than 80 Chinese research papers and patents revealed that researchers linked to the Chinese People's Liberation Army used outputs from American AI models to train local systems for military purposes. The practice relies on a technique known as "model distillation," which allows using the answers of large models to train smaller, cheaper models without the massive computing power needed to develop the original models. The documents indicate the technique was used to develop systems for image analysis and target recognition, and to run military applications on limited-capability platforms such as drones. These developments come as the United States continues to impose restrictions on exporting advanced AI chips to China. Another review of Chinese military procurement requests showed continued demand for advanced American chips such as Nvidia's "A100," alongside a shift toward local alternatives such as Huawei's "Ascend" processors. Chinese researchers are also benefiting from "DeepSeek" models to reduce model size and computing resource consumption to run them on less advanced processors. Despite the progress, challenges remain regarding the reliability of the systems and their ability to operate in complex field conditions, in addition to concerns about data security and the potential for model manipulation.