US agencies accuse six Chinese firms of industrial-scale distillation; Beijing rejects the claims

A joint advisory from CISA, the NSA, and the FBI says six Chinese AI companies have conducted industrial-scale distillation attacks on American frontier models since at least late 2024. The advisory names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, and says they extracted billions of tokens through millions of requests to models from Anthropic, OpenAI, Google, and xAI. The agencies assess that the scale and sophistication of the operations indicate Chinese government awareness, and describe the approach as likely a core development strategy for the firms. Distillation itself is a legitimate technique in which a student model learns from a well-trained model's outputs. The advisory describes how the firms allegedly distributed API requests across fraudulent or shared accounts, cloud services, aggregators, and transfer-station proxies to bypass geographic restrictions, usage limits, and detection. Google had warned in February that distillation attacks can abuse API access to replicate powerful models at a fraction of the training cost. China's Ministry of Commerce responded on Wednesday, saying the allegations lack factual and legal grounds. The ministry described distillation as neutral and widely used by companies worldwide, including American firms, and accused Washington of using the claims to assert technological dominance, monopolise computing power, and suppress competition. It warned that China would retaliate if the allegations were used as a pretext to contain its AI sector. The dispute lands as one of the named companies, DeepSeek, released a new model the same day.