China Claims US Firms Distil Its Models Too, As AI Trade War Rhetoric Escalates

Beijing has warned it will 'take all necessary measures' if Washington moves to sanction Chinese AI companies, firing back at US allegations that Chinese firms have been distilling American frontier models.

The threat landed at a Monday press briefing from China's Ministry of Commerce, triggered by comments from Michael Kratsios, Trump's science and technology adviser, who suggested Chinese companies had been using distillation to essentially piggyback on American AI research. US Treasury Secretary Scott Bessent backed that position.

China's rebuttal follows a familiar script: the allegations are propaganda, and the theft narrative is designed to discredit Chinese innovation rather than reflect reality. The Ministry then adds a more specific defence: Chinese AI companies simply didn't have enough time to distil the latest US frontier models, because American labs published those models only days before Chinese firms released their own work.

Beijing also claims some Chinese models are already world-leading, which creates a curious tension in its own argument. China is simultaneously saying its AI is too good to need distillation, and also that there wasn't enough time to do it. Pick one.

The more interesting counterpunch is the Ministry's allegation that 'many American AI companies have distilled Chinese models in their research and training.' That's actually plausible. Kratsios himself acknowledged last week that distillation isn't inherently dodgy. When the goal is producing smaller, more efficient models, it's a standard and legitimate technique. With a large number of Chinese open source models freely available, US researchers using them wouldn't be doing anything improper.

The Ministry stops short of spelling out what retaliatory measures China might actually take. Instead, it ends with an appeal for cooperation and a nudge toward WAICO, the World Artificial Intelligence Cooperation Organization, a 29-member body China recently convened to promote AI governance. Beijing has also been pushing for AI models to be open sourced globally, framing it as a move toward shared innovation.

That last point puts China in some unexpectedly familiar company. Back in the US, 77 tech companies recently signed an open letter urging lawmakers to protect open weights models. China's government and a significant chunk of Silicon Valley are, at least on that particular question, reading from the same page.

The broader trade row over AI is only getting noisier. Whether any of this posturing translates into actual policy consequences remains to be seen.