Days after Chinese startup Moonshot AI released Kimi K3–world’s first open-weight 3T AI model, the US government has now accused the company of using “distillation techniques” on Anthropic’s frontier AI model Fable to train its own model.
Distillation refers to a machine learning technique where a smaller model is trained to mimic the behaviour and output of a “teacher” AI model. Back in February Anthropic AI had accused three Chinese AI companies–Deepseek, Moonshot and Minimax of distillation; however there has not been any recent direct accusations against Moonshot’s Kimi K3 by the company.
Also Read: What is Knowledge Distillation?
Michael Kratsios, US President Donald Trump’s Science and Technology adviser, was the first official to levy the “theft” allegations on Moonshot AI as he took to X to claim that the company had developed a “sophisticated internal platform” to conduct large scale distillation against US models.
“We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model. 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 access to avoid detection. Moonshot AI has also acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models,” wrote Kratsios on X.
We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.
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…
— Director Michael Kratsios (@mkratsios47) July 22, 2026
Kratsios stressed in his tweet that the US supports the free and fair development of AI, with a competitive ecosystem encompassing frontier models, specialised systems, open-source frameworks, and open-weight models.
“Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem. However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable,” he further expressed.
Similar sentiments were raised by U.S Treasury Secretary Scott Bessent who wrote on X, “We support open-source AI and the innovation it unlocks. But open source is not open season on American IP. When PRC firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table.”
We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.
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…
— Director Michael Kratsios (@mkratsios47) July 22, 2026
The launch of Kimi K3 has set alarm bells ringing in the US policy corridors as the model surpassed its American counterparts Claude Fable and GPT 5.6 Sol in coding and other agentic areas.
The world’s first open weight 3T class AI model was launched on July 16, the same day Chinese President Xi Jinping gave the clarion call of China leading the Global South in AI at the inaugural of World Artificial Intelligence Conference (WAIC) 2026 in Shanghai.
Ever since its launch, multiple independent assessments by researchers including Arena ai have shown Kimi K3 showing frontier-level performance in long-horizon coding, agentic workflows, knowledge work and deep reasoning.
However, on July 24, UK based AI Security Institute and the US Center for AI Standards and Innovation (@NIST), claimed in a fresh report that Kimi K3 scored lower than frontier models of US in cyber capabilities in their preliminary cyber evaluations.
Internet Points Out The Flaws in Distillation Accusation
However, not everyone agrees with the distillation accusation put against Moonshot AI by the US government.
Hours after the tweet by Kratsios went live, users were quick to point out the hypocrisy in AI companies accusing each other of “theft” i.e. illicit distillation technique, while their foundation models were trained on works of millions of real persons on the internet.
Also Read: With Anthropic, AI Companies are Finally Realizing How It Feels to be Scraped
There have been several ongoing copyright lawsuits filed against AI companies by media houses, authors and performance artists alleging that the companies profiteer by simply lifting their work to train their LLMs without any compensation to the original creators.
Alexander Green, founder of Little Bird AI took to X and wrote, “I’ve seen no greater irony in my entire life than Anthropic complaining about distillation. They trained on the entire corpus of human knowledge and are now selling it back to us, compressed. All while desperately trying to game regulators into providing them with a business model and claiming the moral high ground, haughtily and with great condescension.”
I’ve seen no greater irony in my entire life than Anthropic complaining about distillation. They trained on the entire corpus of human knowledge and are now selling it back to us, compressed. All while desperately trying to game regulators into providing them with a business…
— alexander green (@alexframegreen) July 22, 2026
While there has not been any official communication from Moonshot AI on the accusations put by the US government, South China Morning Post (SCMP), China’s leading media platform, quoted a Moonshot employee ridiculing allegations of distillation.
