China Calls US Sanctions Threat 'AI Hegemonism.' Neither Side Has Shown Its Evidence.
Five days after Treasury Secretary Scott Bessent put sanctions "on the table" over the Kimi K3 distillation accusation, China's Ministry of Commerce issued its first official government response on July 27 — dismissing the claim as lacking "any actual evidence," warning that Beijing would protect its open weights llm developers, and calling it a classic case of hegemonic behavior.
The largest open weight llm: What's new since the last update
The prior piece on this dispute left off with Treasury's threat, researcher skepticism about the timeline, and Jensen Huang's public break from the administration's framing. Two things have changed since then.
First, this is no longer just a US accusation sitting unanswered — it's now a formal state-to-state exchange. MOFCOM's statement did two things at once: it rejected the premise that Chinese AI development is parasitic, pointing to the pace and independence of recent Chinese model releases, and it turned the accusation around, stating that "many US AI companies have distilled from China's models for their own R&D and training." The ministry also invoked the roughly 200 US startups that had separately urged Washington not to cut off access to Chinese open-weight models, calling on the US to "listen attentively to objective, rational voices in the industry."
Second, Kimi K3's full weights went public on Hugging Face on July 26 — a day ahead of Moonshot's own schedule — which means outside researchers can now inspect the model's architecture directly rather than relying on API-based behavior alone. That doesn't resolve the distillation question, but it does open a path to independent verification that didn't exist a week ago.
The evidence gap hasn't closed — on either side
The core problem with this dispute hasn't changed: neither government has shown its work. The White House hasn't published the methodology behind Bessent's claim that officials found "watermarks of US large language models" embedded in Chinese systems. China's MOFCOM statement doesn't name Moonshot specifically, doesn't address the 15-day timeline between Fable 5's July 1 return to availability and Kimi K3's July 16 launch, and doesn't rebut the specific Nvidia GB300 chip-routing allegation — it responds to the broader accusation, not the particulars.
There's a wrinkle worth being precise about: Anthropic's underlying attribution — the 3.4 million Moonshot-linked exchanges — comes from a campaign the company documented in February, months before Fable 5 existed. That's real, disclosed evidence of something, but it predates the specific claim in dispute and doesn't by itself establish that Kimi K3 was built from Fable 5 outputs generated in a 15-day window. Both governments are, in effect, arguing past each other: Washington's public framing centers on IP theft via distillation, which is legally murky under current US copyright law since AI outputs generally aren't copyrightable; the more concrete legal exposure — if the Nvidia chip-routing allegation through Thailand holds up — is an export-control violation, a different and more straightforward claim that neither side's public statements have focused on.
Where this sits as of July 29
No formal action has been taken. The Commerce Department's Bureau of Industry and Security is reportedly investigating, but Entity List designation — the mechanism Bessent named — would need that investigation to conclude first. A US-China AI-focused meeting is reportedly scheduled for September, which most reporting suggests makes near-term sanctions unlikely barring a sharper piece of public evidence from either side.
For now, this remains what it's been since July 22: a serious accusation with real diplomatic consequences, resting on evidence that neither government has made available for anyone outside it to check.
