Dilara Hamit
23 July 2026•Update: 23 July 2026
A senior White House official accused Chinese artificial intelligence startup Moonshot AI of conducting a “large-scale” campaign to extract capabilities from leading US models, escalating tensions between Washington and Beijing on advanced technology.
Michael Kratsios, director of the White House Office of Science and Technology Policy, said Wednesday that the US government had information indicating Moonshot used Anthropic’s Fable model to help develop its new Kimi K3 system.
Kratsios alleged that the company created a sophisticated internal platform to carry out model “distillation” while rapidly switching between different methods of access to avoid detection.
Distillation is a common technique through which a less powerful AI system learns from the responses generated by a more advanced model. Although the method has legitimate uses, US officials argue that conducting it covertly and at an industrial scale can amount to intellectual property theft.
“Large-scale, covert industrial distillation to steal American IP is not,” said Kratsios, distinguishing the alleged activity from legitimate research.
He also alleged that Moonshot acquired servers equipped with Nvidia’s GB300 processors and accessed similar hardware in Thailand, potentially to train its models.
The GB300 is part of Nvidia’s advanced Blackwell computing platform. Washington has restricted Chinese access to cutting-edge AI chips because of concerns that the technology could support military and surveillance capabilities.
Moonshot has not publicly responded to the allegations. China’s government has previously rejected similar US accusations, saying the country’s AI advances resulted from domestic research, investment and international cooperation.
Kimi K3 attracted global attention after being unveiled last week, with Moonshot claiming that the model could compete with leading systems developed by Anthropic and OpenAI. Its full weights are expected to be released July 27, allowing developers to download and customize the model.
The system has performed strongly in some coding and agent-based benchmarks, although many of Moonshot’s performance claims cannot be fully independently assessed until the weights are released.
The White House allegations came one day after Treasury Secretary Scott Bessent said the Trump administration was examining whether Chinese AI models had been developed using intellectual property taken from US competitors.
Bessent warned that sanctions and placement on the Commerce Department’s Entity List would be considered when foreign companies crossed the line into intellectual property theft.
“We support open-source AI and the innovation it unlocks. But open source is not open season on American IP,” he said.
Anthropic previously accused Moonshot of using fraudulent accounts to generate more than 3.4 million exchanges with its Claude models. The company said the alleged operation targeted capabilities, including coding, data analysis, reasoning, tool use and computer control.
Some industry experts, however, have questioned whether the recently released Fable model could have been the main source of Kimi K3’s capabilities, given the short period between the two models’ launches.
The controversy comes as Washington intensifies scrutiny of Chinese AI companies and seeks to prevent advanced American chips and proprietary technology from reaching strategic competitors. It also risks adding another source of tension ahead of an expected meeting between US President Donald Trump and Chinese President Xi Jinping in September.
