Washington is calling out China’s AI industry by name. A joint advisory from the NSA, CISA and FBI says labs such as DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI have conducted industrial-scale distillation against US frontier models since late 2024, harvesting billions of tokens from Anthropic Claude, OpenAI GPT, Google Gemini and SpaceXAI Grok with apparent state backing.
The document frames the practice as aggressive and malicious rather than routine research. Campaigns extract chain-of-thought reasoning, fail over automatically when providers block a route, and run quality checks designed to spot defensive countermeasures. The agencies describe the cheap bulk access as flowing from premium subscriptions bought at scale and passed around developer teams.
Specifics include DeepSeek mining reasoning and domain capabilities for R1 and V3 in 2024 and 2025, Moonshot pulling Claude data for Kimi-K3 and GPT-4o output for Kimi-K2, and Alibaba distilling several flagship models. The payoff, per the advisory: shorter development timelines and sharply lower training spend.
Distillation is legal and common in AI research, the agencies concede; what alarms them is the scale, the targeting of restricted capabilities and the government alignment. They advise US model makers to enforce terms of service, watch subscription abuse patterns and build resistance to extraction into their models.
