The US has accused Chinese AI startup Moonshot AI of stealing technology from US AI company Anthropic's most advanced large language model, Fable, during the development of its latest model Kimi K3, and of acquiring advanced Nvidia AI chips. The US has further warned that Moonshot AI may be added to the trade blacklist.
Kelvin Droegemeier, senior technology official in the Trump administration and Director of the White House Office of Science and Technology Policy, posted on social platform X on Wednesday: "Information we have shows that Moonshot AI conducted distillation of Anthropic's Fable model during the development of the K3 model."
Model distillation refers to using a more expensive AI model to train a smaller model, in order to reduce the training costs of developing high-performance AI systems.
Droegemeier accused the company of secretly carrying out industrial-scale model distillation to steal US proprietary technology and undermine US scientific research achievements, calling such behavior "unacceptable."
He also said that Moonshot AI obtained servers equipped with Nvidia GB300 AI chips and used such servers in Thailand, "very likely for training its AI models."
On the same day, US Treasury Secretary Besant also posted on X that Moonshot AI may be considered for inclusion on the trade blacklist and subject to company sanctions.
In response, Chinese Embassy in the US spokesperson Liu Chang stated that Droegemeier's accusations are "groundless", and emphasized that China respects intellectual property protection.
He said: "Relevant US officials should respect the facts, abandon prejudice, and stop smearing and slandering the development achievements of China's AI industry."
The latest allegations from the Trump administration are expected to further intensify US-China tensions over AI technology. Currently, companies in both countries are launching increasingly powerful AI models, which not only have broad economic impact but may also have military applications.