In a dramatic development that should surprise no one in hindsight, a high-level official in the Trump administration has just declared that Moonshot's Kimi K3 model has been distilled from Anthropic's Fable, and further trained on NVIDIA's GB300 GPUs in Thailand.
Michael Kratsios asserts that Moonshot covertly owns some NVIDIA GB300 servers, and accessed additional GB300 GPUs in Thailand to train its Kimi K3 model
As we reported recently, the proclivity of China's AI labs to distill frontier US models is raising hackles within the Trump administration, especially given the fact that there is no easy way of preventing this. After all, Chinese engineers often take their model-laden hard drives to US-friendly countries to imbue their models with frontier-level capabilities by using reinforcement learning techniques on cutting-edge NVIDIA GPUs.
Of course, it was already clear that Moonshot's viral Kimi K3 model had been distilled from Anthropic's Claude. After all, the Kimi K3 was caught identifying itself as "Claude, an AI assistant made by Anthropic" in at least one conversation.
Now, however, the current Chief Science and Technology Advisor and the Director of the White House Office of Science and Technology Policy (OSTP), Michael Kratsios, has jumped into this melee, openly declaring that the US government has "information" that Moonshot distilled its Kimi K3 model from Anthropic's Fable.
According to Kratsios, Moonshot "developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection."
What's more, the high-flying Chinese AI lab apparently already owns some NVIDIA GB300 servers, and was able to access additional GB300s via Thailand.
Apparently, Moonshot has been able to evade detection by rapidly rotating between a number of access methods, while continuously querying Anthropic's models. What's more, Moonshot's ability to leverage NVIDIA's GB300 GPUs only made its distillation more efficient.
This is not to say that Moonshot did not add any value. After all, the Kimi K3 model is quite efficient to run on data center-scale hardware, courtesy of its dramatically reduced KV cache and built-in optimizations for per-token memory use - achieved by distributing its 896 experts across a large number of GPUs.
Note: The article was updated to replace the words "solid evidence" with the word "information," in accordance with Kratsios' re-characterization.
About the author: Writing is my one incontrovertible passion. Over the past six years, he has authored over 2,200 distinct articles on financial and tech-related topics, spanning nearly 1 million words. And he has been a member of Wcctech mobile team since 2025. As an alumnus of the University of Toronto, Rotman Commerce Program, I bring nuance, in-depth knowledge, and a unique perspective to every topic that I cover. When I'm not writing, I'm traveling the world, exploring hidden confectionaries and restaurants as an aspiring food connoisseur.
Follow Wccftech on Google to get more of our news coverage in your feeds.
