The White House has accused China’s Moonshot AI of conducting a “massive” campaign to steal functionality from Anthropic’s top AI models through a technique called distillation.
In a post on X on Wednesday, President Trump’s science and technology advisor Michael Kratsios claimed that Moonshot extracted the features of Anthropic’s Fable AI model to develop the newly announced Kimi K3 model.
Treasury Secretary Scott Bessent said Wednesday that sanctions “will be on the table” if Chinese companies “cross the line” into intellectual property theft through “industrial-scale distillation attacks.”
What is distillation?
Distillation is the process by which a weaker AI model queries a stronger AI model to extract its features, essentially reverse engineering advanced features through systematic queries. Kratsios’ claims suggest that Moonshot used this technique to compress Anthropic’s Fable AI technology into a competitive product.
Bessent made the bet clear on social media, saying, “We support open source AI and the innovation it unleashes. But open source is not open season on American intellectual property.”
Kratsios also said that Moonshot illegally accessed Nvidia’s advanced servers and used Nvidia’s GB300 Grace Blackwell computing system to train its models. This claim is important.
The US government had restricted exports of Nvidia’s cutting-edge chips to China in 2022 due to national security concerns. Despite strict measures against illegal trade in chips being implemented globally, such activity remains difficult to detect.
Moonshot AI Kimi K3
Moonshot plans to launch the Kimi K3 as an open source model by July 27th, and that’s right around the corner. Once a model starts as an open source model, it becomes difficult in a legal sense to sue for IP theft.
Since its launch last week, the Kimi K3 has been turning heads around the world. According to many, the United States has succeeded in bridging the gap in AI model capabilities between the West and China, which is exactly what the United States wants to avoid.
This is not the first time a company has come forward with such accusations. Just two months ago, in June, Anthropic accused Chinese online commerce platform Alibaba of illegally extracting the capabilities of Claude AI through a distillation attack.
