The Trump administration has formally accused Chinese startup Moonshot AI of stealing intellectual property from Anthropic's most advanced model, Fable, to develop its newly released Kimi K3 system. The allegations, made by White House Office of Science and Technology Policy Director Michael Kratsios, mark a sharp escalation in the technology rivalry between Washington and Beijing.

Kratsios said the administration has information that Moonshot AI distilled Anthropic's Fable for the development of its K3 model. He described a 'sophisticated internal platform' used for 'large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research.'

“Large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.”

Treasury Secretary Scott Bessent said on Wednesday he was considering adding Moonshot AI to a trade blacklist and imposing sanctions. Bessent had earlier told Fox Business that Washington could sanction Chinese labs for intellectual-property theft, saying 'we are finding watermarks of our U.S. large language models on many of the Chinese models.' The Commerce Department's Bureau of Industry and Security is formally investigating Chinese firms like Moonshot over their use of restricted chips, a spokesperson told The Information.

The chip allegations

Kratsios also claimed Moonshot AI acquired servers containing Nvidia GB300 chips and accessed them in Thailand, 'likely to train its AI models.' The GB300 is part of Nvidia's Blackwell line, which the US bars from sale to Chinese entities. A May memo reaffirmed the ban on selling Blackwells to Chinese companies. Nvidia did not respond to requests for comment.

Kimi K3 and the market reaction

Moonshot released Kimi K3 last week. The open-weight model quickly climbed leaderboards, with Moonshot claiming it trails only Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 on overall capability. Its arrival set off a rout in tech stocks, echoing the DeepSeek shock from early 2025. The Financial Times reported that Kimi K3, along with other recent Chinese models like GLM 5.2 and Alibaba's Qwen 3.8, have 'all but erased the time advantage' US frontier labs once held.

“For 98% of customers, the 1 to 2% difference isn't worth paying 10 to 20 times more.”

Skepticism from AI experts

Several AI researchers questioned whether distillation alone could explain Kimi K3's capabilities. Braden Hancock, a researcher at the Laude Institute and co-founder of Snorkel AI, told TechCrunch the timeline does not support the claim. Fable has only been publicly available since July 1, and Hancock said 'you can't distill that much data, train a model, and release it in two weeks.' Nathan Lambert, an AI researcher at the Allen Institute for AI, said distillation has become 'less and less impactful over time as the Chinese models get closer to the frontier.' Even a top OpenAI executive conceded that Kimi K3's advances did not look like the result of copying, according to the Financial Times.

“I don't think you get a model this strong and this quickly on the heels of Fable doing strictly distillation. There's just not even frankly time, right?”

Divisions inside the White House

POLITICO reported a split within the administration. Kratsios and Bessent are taking a hardline approach aimed at protecting American IP, while Commerce Secretary Howard Lutnick has a more moderate stance, concerned with companies' access to cheaper technology. In the six days since Kimi K3's release, Bessent and Kratsios offered diverging policy ideas, none of which have been publicly announced by the White House. Press Secretary Karoline Leavitt confirmed 'these conversations are taking place.' Reports that the administration was considering banning Chinese open-weight AI models caused widespread panic in Silicon Valley, but two people familiar with the discussions told POLITICO no such sweeping proposal was seriously on the table.

Responses from Beijing and Anthropic

Chinese embassy spokesperson Liu Chang called Kratsios' comments 'entirely unfounded,' adding that China respects intellectual property protections. Foreign ministry spokesperson Lin Jian said such actions 'will only disrupt the development process of global artificial intelligence.' Anthropic public policy executive Sarah Heck said Chinese theft of US models 'creates serious national security risks for the United States.' Anthropic had earlier sent letters to US lawmakers formally accusing Moonshot, DeepSeek, and MiniMax of industrial-scale distillation of its Claude models.

“Chinese theft of U.S. models creates serious national security risks for the United States.”