The United States government accused Chinese company Moonshot AI of allegedly stealing trade secrets related to artificial intelligence.
According to AFP, U.S. authorities claim that the Asian firm extracted capabilities from the Fable model of U.S.-based Anthropic to develop its own Kimi K3 system. This statement, published on social media platform X by Michael Kratsios, director of the White House Office of Science and Technology Policy, escalated the technological dispute between Washington and Beijing.
The conflict revolves around the technique of "distillation," a method widely used in the industry to train smaller models based on responses from advanced systems. Although this is a common practice, U.S. officials assert that Chinese companies use it on a massive scale to replicate protected American systems.
According to Kratsios, Moonshot allegedly built a sophisticated internal platform that facilitates large-scale distillation against U.S. models, allowing it to alternate between various access methods and evade detection systems.
Concern within the U.S. administration grew after discovering that Moonshot AI had allegedly acquired servers equipped with Nvidia GB300 chips, technology whose export to China is restricted by U.S. controls.
Kratsios specified that the company also managed to access servers with that hardware in Thailand, presumably to train its AI models, suggesting a potential mechanism to bypass the limitations imposed by Washington. According to the White House, this maneuver points to Chinese companies seeking competitive advantages in global AI development.
In statements aired on Fox Business, Treasury Secretary Scott Bessent warned that sanctions could be imposed on foreign companies involved in appropriating U.S. technology. "If we see, especially, foreign models stealing from our big companies, we have the ability to sanction them for that theft," the official said.
Bessent emphasized that authorities have identified "watermarks of our American models in many Chinese models, and that is unacceptable."
The accusation against Moonshot comes amid growing rivalry between the United States and China in the AI sector. The controversy intensified after the Kimi K3 model from the Asian company raised concerns due to its technological proximity to developments by firms such as Anthropic and OpenAI.
Anthropic revealed that Moonshot used its models more than 3.4 million times, violating terms of service and geographic restrictions that prohibit access from China. The Fable model, allegedly used in training Kimi K3, debuted in June, indicating that the activity may have continued until recently.
The case also sparked a debate on AI development models. While the United States predominantly uses "closed" systems with restricted access, China promotes an "open source" approach. Jacob Helberg, an official linked to the Trump administration, described Moonshot's actions as "theft of invaluable intellectual property from the United States."
He argued that "lying, cheating, and stealing is not innovation or cooperation," referring to China's official discourse on mutual benefit in technology. Meanwhile, Dean Ball, an executive at OpenAI, stated that Chinese open-source models represent progress toward what he called "total artificial intelligence communism."
The rising tension coincides with the White House's concern over the advancement of AI systems capable of executing sophisticated cyberattacks.
