Trump science adviser Michael Kratsios accused Moonshot AI of distilling Anthropic's Claude to build K3.
Why it matters: The allegation puts a leading Chinese AI lab under sharper U.S. scrutiny over how it may use American models. It could also intensify attention on AI model training practices and access to advanced chips.
- Michael Kratsios said Moonshot AI distilled Anthropic's Claude for its K3 model, Reuters reported on July 22, 2026.
- Anthropic said on Feb. 23, 2026 that it found industrial-scale distillation campaigns by DeepSeek, Moonshot and MiniMax.
- Anthropic said those campaigns involved more than 16 million exchanges with Claude through about 24,000 fraudulent accounts.
- Reuters later reported Anthropic said a separate campaign involved more than 28.8 million exchanges, including more than 3.4 million tied to Moonshot.
White House science and technology adviser Michael Kratsios accused Beijing-based Moonshot AI of using Anthropic's Claude model to help develop its K3 model, according to Reuters.
Kratsios said, "We have information that Moonshot AI distilled Anthropic's Claude for the development of its K3 model," Reuters reported. The allegation follows Anthropic's public statements this year about what it called large-scale distillation activity by Chinese AI firms.
In a Feb. 23 statement, Anthropic said it had identified industrial-scale campaigns by DeepSeek, Moonshot and MiniMax, involving more than 16 million exchanges with Claude through about 24,000 fraudulent accounts, according to Anthropic.
Reuters later reported that Anthropic said a separate campaign by Chinese firms involved more than 28.8 million exchanges with Claude between April 22 and June 5, 2026, and that Moonshot was implicated at a scale of more than 3.4 million exchanges. Anthropic said distillation attacks at that scale require a coordinated response across the AI industry, cloud providers and policymakers.
The Reuters report did not say Kratsios announced sanctions or export controls. It did show the accusation landing amid broader U.S. concern about Chinese AI development and access to advanced technology.
By the numbers
- 16 million - Anthropic said that many exchanges with Claude were tied to industrial-scale distillation campaigns.
- 24,000 - Anthropic said the exchanges went through about that many fraudulent accounts.
- 3.4 million - Reuters reported Anthropic said that many exchanges were tied to Moonshot in a separate campaign.
Yes, but: The allegation comes from a Reuters report and Anthropic's prior disclosures, but Moonshot's response was not included in the materials provided.