US intelligence agencies, including the NSA, FBI, and CISA, have issued a joint statement accusing several Chinese AI companies of large-scale model distillation. The agencies claim that DeepSeek, Moonshot AI, Alibaba Group, MiniMax, StepFun, and Z.ai have been regularly distilling American LLMs since at least 2024 to accelerate the training of their own systems.
The report details how specific Chinese firms used American models to improve their AI's ability to solve mathematical problems and analyze software code. US officials believe these activities were carried out with the knowledge of the Chinese government and involved the intentional violation of usage restrictions on US models.
In response, the Chinese embassy in the US called the accusations a deliberate attack on Chinese developers and AI progress, stating that such politicization harms the global industry. Meanwhile, US intelligence has urged American developers to strengthen protection measures and increase data sharing to detect such activity.
The US government is currently considering measures to stop this practice. US Treasury Secretary Scott Bessent noted in July that violators from the Chinese side could face American sanctions.
