An advisory that names without binding
Three agencies signed the text on Tuesday: the FBI, the NSA and CISA. It says DeepSeek, Alibaba and Moonshot AI distilled billions of tokens from US models since at least late 2024, distillation being the training of a lesser model on a more advanced one, and it places that method at the core of their model development. All of this is a finding. None of it is an obligation: the advisory carries no licence condition, no export restriction and no penalty a company could be made to answer for.[1]
China's Commerce Ministry answered on Wednesday with a sentence built the same way. If the United States suppresses Chinese AI companies under the pretext of targeting distillation, the ministry said, China will take resolute countermeasures. The threat depends on a US action that has not been taken, and it names no measure of its own. Foreign ministry spokesperson Mao Ning kept to the same register, asking Washington to refrain from unfounded accusations.[1]
Where the contest actually sits
The two statements argue about a definition. The Commerce Ministry says distillation is commonplace in the industry, including among US companies. The advisory says the scale and sophistication of these campaigns move it somewhere else. Whoever fixes that definition first also decides whether a later measure reads as enforcement against theft or as protection of an incumbent.[1]
Treasury Secretary Scott Bessent gave a third description in Dallas on Tuesday: the Chinese distill American models and can never get ahead. That framing treats distillation as a ceiling on a rival rather than a wrong needing a remedy, and a ceiling calls for no instrument at all. Three descriptions of the same conduct, and no agreed name for it.[1]
The summit sets the clock
The advisory landed weeks before Xi Jinping is due in the United States for a summit with Donald Trump that is to cover AI, and according to Reuters he could travel with a large number of Chinese executives. China is pouring nearly 300 billion dollars into artificial intelligence and robotics. A public finding issued now is already on the table for that meeting, whether or not anyone converts it into a rule.[1]
The measurable question is whether the finding turns into a restriction. Should the United States name DeepSeek, Alibaba or Moonshot AI in a measure that limits their access before the end of 2026, the advisory does the work an export control usually does, and the conditional threat from Beijing finds its trigger. Absent such a measure, the advisory works as a marker laid on the table ahead of a negotiation.[1]