China's Ministry of Commerce published a spokesperson's response at 18:13 Beijing time on 27 July — 10:13 UTC — addressing reports that Washington intends to investigate and sanction Chinese AI companies over the distillation of American frontier models. The ministry did not deny the practice. It said the Americans do it too.
The inversion
The key line, in the original: "It is understood that many US artificial intelligence enterprises have distilled Chinese models in research, development and training." The ministry argues the accusation is implausible on its own terms, noting that Chinese and American model release dates now sit very close together and that some Chinese models already lead — which would make one-directional copying an awkward fit with the timeline.
The language
The statement calls the American conduct "typical AI hegemonism" and warns that "for any conduct that substantially harms China's interests, China will take all necessary measures." It also reaches for an American constituency, citing nearly 200 US startups that have urged Washington not to cut off access to Chinese open-source models.
What was not announced
Coverage framing this as retaliation gets it wrong twice. No countermeasure was announced — this is a press question-and-answer, not a rule, an entity listing or a tariff. And Beijing is responding to statements by American officials, not to a promulgated regulation: no US investigation has been formally opened on the public record. "All necessary measures" is a conditional warning about a future that has not happened yet on either side.
Why it is worth reading anyway
Because it is the factual premise, not the rhetoric, that will matter. Any future Commerce or Treasury designation over distillation will have to survive a record in which the target government has formally asserted that the practice runs both ways. Distillation is also genuinely hard to prove: it leaves statistical traces in model behaviour rather than documents, which is precisely why the same accusation is now travelling in both directions with roughly equal confidence and roughly equal evidence.
