A top White House official has accused a Chinese startup of covertly stealing technology from American artificial intelligence company Anthropic and accessing restricted Nvidia chips to power its newest model, with the Trump administration now warning sanctions could follow.
Michael Kratsios, director of the White House Office of Science and Technology Policy, said Wednesday the administration has confirmed that Moonshot AI, a China-based firm, distilled Anthropic’s Fable model to produce Kimi K3, the company’s latest AI release. Distillation is a technique in which the outputs of a more powerful model are used to train a lesser one, effectively copying its knowledge.
“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,” Kratsios wrote on X.
Kimi K3 was released late last week and immediately rattled the AI industry. The model claimed performance on par with Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 Sol, particularly in coding tasks.
Kratsios also charged that Moonshot obtained servers equipped with Nvidia’s GB300 system, including access to the same hardware in Thailand. The GB300 contains Nvidia’s advanced Blackwell chips, which are prohibited from export to China. The highest chip currently licensed for China is the H200.
Treasury Secretary Scott Bessent said Tuesday the administration is preparing to act.
“We are finding watermarks of our U.S. large language models on many of the Chinese models, and that’s unacceptable,” Bessent told Fox Business’s Maria Bartiromo. “If we see that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft.”
China pushed back Wednesday. Foreign Ministry spokesperson Lin Jian told reporters the country “opposes politicizing and instrumentalizing trade and tech issues” and called the accusations politically motivated. Liu Chang, a spokesperson for the Chinese embassy in Washington, called the allegations entirely unfounded “smear.”
Independent analysts are less certain. Hamish Low, a research associate at the Institute for AI Policy and Strategy, said after reviewing Kimi K3 that it is “pretty overwhelmingly likely they are distilling from leading U.S. models.” He noted the model excels in coding, where distillation data is more accessible, but falls short in cybersecurity, which U.S. firms guard more heavily.
The Trump administration flagged similar behavior in April, when Kratsios said Chinese entities had used thousands of proxy accounts and jailbreaking tactics to access proprietary U.S. AI information. Bessent’s sanction warning Tuesday was the sharpest public threat yet.
Kratsios distinguished between legal and illegal use of the technique. “Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem,” he said. “However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.”
