Washington Accuses Chinese AI Firms of Systematic ‘Distillation’

The US government formally accused Chinese artificial intelligence companies of unauthorised distillation of American models.^2,5 Officials said China-based AI companies are “conducting systematic extraction of proprietary functionalities” from those systems.² More specifically, Washington said Chinese firms siphon the outputs of models built by OpenAI and Anthropic in order to train their own.⁴

China hit back against the claims, rejecting the allegations of malicious distillation on 9 September.^3,6 The exchange comes ahead of planned talks between Trump and Xi.^3,6 Neither side has stepped back from its position, and the accusation is being carried into the meeting rather than resolved before it.^3,7

Why it matters to investors

The accusation ties directly to existing chip policy. With computing resources reduced by export controls on advanced AI semiconductors, Chinese corporations turned to the outputs of US models as an input.^4,5 On that account distillation is the workaround for a hardware restriction, which makes it an enforcement problem rather than evidence the controls failed.^4,5 It also implies the next control point is model access rather than chip supply.^2,4

The legal ground is unsettled. The Wall Street Journal reported that the output of AI models is unlikely to be considered protectable in the way other assets are.¹ That gap matters for anyone holding US model developers, because it determines whether this becomes a trade-policy fight or a litigable one.^1,2 Framed as US AI industrial policy, the complaint belongs to the government rather than to the companies whose outputs are said to have been taken.²

What to watch

The rhetoric is escalating into the meeting rather than softening ahead of it.⁷ Semafor reported the US Treasury secretary warning that “nothing else would matter” if China were to win the AI race, which is about as high as an official can place the stakes.⁷

Two markers are worth tracking. The first is whether the distillation charge produces a concrete enforcement action, given the industrial-policy framing rather than a single company’s complaint.² The second is whether the Trump-Xi meeting proceeds on schedule with the accusation and Beijing’s rejection both on the table.^3,6 Coverage outside the United States is already treating the dispute as a widening of the US-China AI race rather than a discrete legal matter.⁶ Korean reporting of the same announcement placed the emphasis on the export-control context, describing Chinese firms as turning to American model outputs precisely because advanced chips were restricted.^4,5 That reading has a direct consequence for policy design: if the constraint on compute pushed the behaviour toward outputs, then tightening compute further does not obviously address it.^4,5 The named developers whose outputs are said to have been siphoned, OpenAI and Anthropic, are the parties with the commercial exposure and the least clear legal remedy.^1,4