China Released a 2.8-Trillion-Parameter Model for Free: America Called Moonshot a Thief and Provided No Evidence other than “We Have Information.”
No evidence again.
White House science adviser Michael Kratsios just accused Moonshot AI of a “large-scale” industrial distillation campaign to steal Anthropic’s Fable and build Kimi K3. He also claimed they somehow got hold of banned Nvidia GB300 servers in Thailand.
His entire case? “We have information.”
That is it. No logs, no query volumes, no technical artefacts, no chain of custody and no independent verification. Just the same phrase American officials reach for every time a Chinese company produces something they cannot match.
Even the BBC write-up is a straight repeat of the claim, complete with hedging words like “likely” and the usual “said” and “accused”. They present zero evidence of their own: standard media amplification of an empty accusation.
The timeline alone kills the story completely. Fable became publicly available around 1 July, Kimi K3 previewed two weeks later. Independent researchers, including Braden Hancock, Nathan Lambert and OpenAI’s own Dean Ball, have already said a full frontier-scale distillation in that window is not credible as the primary explanation.
Pre-training a 2.8-trillion-parameter model takes months. You do not magic one into existence by querying an API for a fortnight.
This is the exact same playbook we have watched for years. Huawei “stole” 5G, TikTok “stole” data, semiconductor firms “stole” process technology. Now the open-weight model that just embarrassed the American closed-source stack is “stolen” too. Always the accusation, never the proof.
Distillation is not some exotic Chinese crime: it is standard industry practice. American labs do it. They just do not like it when the student starts outperforming the teacher and then releases the weights for free.
Moonshot has already denied the claim and pointed to its own architecture work. The Chinese embassy called the accusations entirely unfounded.
The actual model drops its full open weights tomorrow. The world will be able to inspect it, run it and decide for itself.
What we are watching is not a serious intellectual-property case: it is the sound of a country that spent years restricting chips, restricting talent and restricting open science discovering that the other side kept building anyway. When you cannot win the race, you try to disqualify the winner.
Same script, same empty evidence, same result.
How many more times do they expect the world to believe “we have information” with nothing to show for it?
(James Wood)
Author: Saikat Bhattacharya
