US Accuses China’s Moonshot of Using Anthropic to Advance AI Capabilities
The US alleges that China’s Moonshot AI employed knowledge distillation to create its K3 AI model using Anthropic’s technology, sparking concerns about intellectual property and AI competitiveness. China has strongly denied the allegations.
Michael Kratsios, director of the US government's Office of Science and Technology Policy (OSTP), has claimed that the K3 model developed by China's Moonshot AI was based on Anthropic's strong Fable model. The open-source K3 model has been creating headlines due to its impressive features and the fact that it is freely available to the public.
This makes it distinct from cutting-edge models developed by American firms like Anthropic and OpenAI. When it comes to artificial intelligence systems, frontier models are at the top of the line, whereas open-weight models allow developers to see how the model is trained. Consequently, user can personalise and manage them separately from the company's service, which is a huge plus. "We have information that Moonshot AI distilled Anthropic's Fable for the development of its K3 model," Kratsios wrote on X.
We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.
— Director Michael Kratsios (@mkratsios47) July 22, 2026
To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of…
How Moonshot Able to Copy Fable: Explained Kratsios
Kratsios detailed the procedure by saying that they built a complex internal platform to distil on a large scale compared to US models. Thus, they are able to swiftly switch between various access methods in order to evade detection. A big, complicated, and computationally expensive model (the "teacher") trains a smaller, quicker model (the "student"). This process is called artificial intelligence distillation or knowledge distillation.
Legal AI distillation to produce smaller, more efficient models is acceptable and essential to this open innovation ecosystem, according to US President Donald Trump's tech advisor. But shady industrial distillation on a massive scale with the goal of stealing American intellectual property and stifling American science is intolerable.
Moonshot AI, according to Kratsios, has purchased servers that run on Nvidia's newest GB300 AI chips. On top of that, it has reportedly trained its AI models by accessing GB300 systems in Thailand. Designed to train and execute complex AI models at scale, the GB300 is Nvidia's newest high-end AI computing platform.
China Refutes US Claims
China denied the claims, stating that its artificial intelligence industry has developed autonomously while adhering to IP laws. The assertions were deemed "utterly unfounded" by Liu Chang, a spokesman for the Chinese embassy in Washington. Consequently, it is recommended that US officials acknowledge China's advancements rather than endeavouring to undermine them. In a separate development, Lin Jian, a spokesman for China's foreign ministry, has lambasted the United States for turning technology competition into a geopolitical struggle.
Further arguing that this will eventually impede global innovation. Claims have surfaced not long after Moonshot introduced Kimi K3, which it called the largest open-weight AI system in the world with 2.8 trillion parameters. Despite being launched just weeks after the US government withdrew Anthropic's Fable and Mythos models over security concerns, Moonshot said the model gives performance close to Fable.
