White House science adviser Michael Kratsios accused Beijing-based Moonshot AI of stealing proprietary technology from Anthropic to develop its Kimi K3 model. The U.S. government alleges Moonshot used industrial distillation
and restricted Nvidia chips to narrow the gap between Chinese and American AI capabilities, prompting threats of sanctions.
While Moonshot AI claims the model can rival top U.S. technology, the Trump administration asserts this parity was achieved through the covert extraction of American intellectual property.
Industrial Distillation of Anthropic’s Fable
At the heart of the dispute is a process known as distillation, where a weaker AI model is trained using the answers generated by a more powerful one. While the practice is common and legal when done openly, Michael Kratsios, Director of the White House Office of Science and Technology Policy, characterized Moonshot’s actions as industrial distillation
aimed at stealing technology.
We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model… However, large-scale, covert industrial distillation aimed at stealing proprietary U.S.
Michael Kratsios, Director of the White House Office of Science and Technology Policy
The scale of the alleged theft is significant. According to BeInCrypto, Anthropic traced more than 3.4 million Claude exchanges to Moonshot in February, utilizing hundreds of fake accounts. Some of these account records reportedly matched the public profiles of senior staff at Moonshot. The campaign specifically targeted Claude’s capabilities in vision, coding, and reasoning.
Circumventing Nvidia Chip Export Controls
Beyond software theft, the U.S. government alleges that Moonshot AI bypassed strict hardware export controls. Kratsios stated that the company gained access to restricted cutting-edge Nvidia servers to train its models. Specifically, BeInCrypto reported that Moonshot allegedly used NVIDIA GB300 servers—chips blocked for sale to China—by operating them in Thailand.
Washington tightened AI chip export clamps in May, and a June raid on a Super Micro office in Taiwan reportedly targeted chip smuggling into China.
Kimi K3 Market Impact and Scale
Its 2.8 trillion parameters allow it to match or rival systems from OpenAI and Anthropic on key benchmarks, particularly in coding tasks.
The “open-weight” nature of the model—meaning its architecture is publicly accessible—creates a unique security risk. Anthropic has warned that copied models often lose their original safety guardrails. Because the full model weights are scheduled for public release by July 27, the technology could spread beyond the control of any single entity.
This development has already rattled investors.
U.S. Policy Response and Sanctions
The Trump administration is treating this as a critical juncture. David Sacks, an AI advisor to the administration, described the situation as an inflection point
for AI policy regarding open-source competition. The administration is now considering placing Chinese AI labs on the Commerce Department’s Entity List.
Treasury Secretary Scott Bessent has signaled that the U.S. is prepared to move from observation to enforcement. Speaking to Fox Business, Bessent addressed the threat of open-source models to U.S.
While the U.S. maintains a 23x spending lead in AI, the rapid ascent of models like Kimi K3 suggests that industrial distillation
may allow rivals to bypass the need for massive capital investment by simply extracting the logic of existing American systems.
The BBC reports that this increased scrutiny comes just before President Trump is expected to meet with Chinese President Xi Jinping in September.
Related reading
