Full Breakdown
U.S. Officials Accuse China’s Moonshot AI of Large-Scale Model “Distillation”
7/23/2026, 11:23:38 AM
Alleged Industrial-Scale Theft of U.S. AI Capabilities
Science and Technology adviser Michael Kratsios said Moonshot AI, a Chinese lab, has used “distillation” – extracting answers from a stronger model – to copy the capabilities of Anthropic’s Claude Fable 5 model for its new K3 system. Kratsios posted that the effort involved “restricted cutting-edge Nvidia servers” and described the campaign as an “industrial-scale” attempt to undermine American research and access proprietary information. Treasury Secretary Scott Bessent warned that sanctions “will be on the table” for Chinese firms that cross the line into intellectual-property theft.
Context of U.S. Export Controls and Chinese AI Push
The United States restricted export of Nvidia’s most advanced chips in 2022 over military-use concerns, prompting worldwide crackdowns on chip smuggling. Despite those limits, Moonshot’s K3 model, unveiled last week, is said to rival leading U.S. systems from Anthropic and OpenAI. In June, Anthropic accused Alibaba of “illicitly” extracting its Claude AI capabilities via similar distillation attacks, urging Congress to strengthen penalties.
Official U.S. Statements
- “We support open-source AI and the innovation it unlocks. But open source is not open season on American IP,” — Scott Bessent, after treasury secretary.
Chinese Government Position
The Chinese government has not issued a new comment on the latest accusations.
Upcoming Open-Source Release
Moonshot AI plans to release Kimi K3 as an open-source model on 27 July, marking the first model of its scale that can be freely downloaded and customized by developers and the public. The scheduled release is expected to intensify the debate over how the United States should respond to foreign AI development that leverages U.S. technology.