SerpApi.Org
The White House has accused one of China’s top AI companies of stealing American technology amid concerns the US is losing its lead in the AI race.
Michael Kratsios, the director of Donald Trump’s Office of Science and Technology Policy, claimed Chinese start-up Moonshot had extracted trade secrets from US company Anthropic on an “industrial” scale.
The intervention escalates a rivalry between the US and China about AI, and could lead to sanctions on Chinese providers.
Moonshot’s new AI model, Kimi K3, has prompted concerns that Chinese AI is closing the gap with leading US companies, coming close to the technology developed by Anthropic and OpenAI.
However, Mr Kratsios said the company’s success had in part been down to China covertly taking advantage of American technology.
He said Moonshot had used a technique known as “distillation” to extract abilities from Anthropic’s powerful Fable system, as well as circumventing a ban on high-end Nvidia chips.
“We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model,” he wrote on X.
“They developed a sophisticated internal platform to conduct large scale distillation against US models, allowing them to quickly switch between multiple methods of access to avoid detection.
“Moonshot AI has also acquired [Nvidia] GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models.”
He added: “Large-scale, covert industrial distillation aimed at stealing proprietary US technology and undermining American research is unacceptable.”
US export controls forbid Chinese companies from obtaining Nvidia’s top chips, even outside the country. The components are seen as a crucial US advantage in the AI race.
Distillation involves developing an AI system using the answers from a more powerful one, allowing it to mimic the powerful system’s abilities. The major AI labs ban the practice, and seek to shut it down when they detect it.
Anthropic said earlier this year that Moonshot had used its models more than 3.4 million times, breaking both its terms of service and geographical restrictions banning Chinese users from accessing it.
Fable, the high-end model allegedly used to train the latest version of Kimi, was only released in June, suggesting the activity had continued since.
Jacob Helberg, a Trump-administration official, called Moonshot’s actions a “heist of invaluable American intellectual property”.
“Lying, cheating and stealing is not innovation or ‘win-win’. It’s corrosive to productive economic activity and to the healthy economic development of nations,” he added.
This appeared to be a reference to Xi Jinping saying that China supported “win-win cooperation” by promoting so-called open-source AI technology, which can be freely downloaded and altered in contrast to leading US “closed” AI systems.
Open source systems also represent a threat to Anthropic and OpenAI’s businesses, which rely on charging premiums for access to what had hitherto been seen as the most advanced AI models.
Dean Ball, an executive at OpenAI, said open source Chinese models were a step towards “full AI communism”.
The White House is reportedly considering stronger controls on Americans using Chinese AI systems.
Concerns have increased in recent weeks because the latest AI systems are capable of carrying out advanced cyber attacks.
Scott Bessent, the US Treasury secretary, said earlier this week that Chinese companies found to have stolen American technology could face sanctions.
“If we see, especially, that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft,” he said.
Moonshot and Anthropic were contacted for comment.
Recommended
China's baby-faced billionaire sends global markets into panic
