The White House said Dark Side of the Moon violated the Distillation Anthropic model. The U.S. Treasury Department threatened sanctions. U.S. Treasury Secretary Scott Bessant further strengthened his previous warning against Chinese artificial intelligence companies on Wednesday, making it clear that if the accusations are true, sanctions will still be considered.

The White House said Dark Side of the Moon illegally distilled the Anthropic model. The U.S. Treasury Department threatened sanctions. U.S. Treasury Secretary Scott Bessant further strengthened his previous warning against Chinese artificial intelligence companies on Wednesday, making it clear that if the accusations are true, sanctions will still be considered. Previously, White House officials publicly accused the Chinese company Dark Side of the Moon of improperly distilling the model of the American artificial intelligence company Anthropic. Model distillation is a common artificial intelligence training technology. Its core principle is to allow smaller models to imitate and learn the results output by larger models. Although this technology is widely used to improve model efficiency, it may also trigger disputes over intellectual property infringement. Bessant said on social platforms that open source does not mean that U.S. intellectual property rights can be violated wantonly. When some Chinese companies carry out covert and industrial-scale distillation attacks and cross intellectual property red lines, sanctions and entity list sanctions will be readily available tools. Prior to this, Bessant had said that the U.S. government would strictly review open source models from China to screen for intellectual property theft and be prepared to impose sanctions if violations are discovered. Bessant's comments came just hours after Michael Kratzios, director of the White House Office of Science and Technology Policy, publicly accused China-based Dark Side of the Moon of mass distilling American models. He accused Dark Side of the Moon of obtaining Nvidia servers equipped with GB300 chips and used them to train its artificial intelligence models by accessing GB300 chips in Thailand. This series of activities raised serious questions about whether it violated U.S. export control rules. It is reported that Nvidia’s GB300 series belongs to the Black Parker architecture products and is currently prohibited from being sold to Chinese companies. In response to this controversy, some industry experts have questioned whether the Kimi K3 model is mainly trained by distilling the Fable model of Anthropic, because Fable was not officially released to the public until July 1. The Dark Side of the Moon officially released the open-source weighted K3 model last week. Its outstanding performance has triggered scrutiny of the original business model of the leading artificial intelligence laboratories in the United States. It has also raised questions about whether these laboratories can continue to support the huge capital investment required for cutting-edge artificial intelligence competitions. The incident further intensified the controversy in Washington politics surrounding the influx of open source models from China. Some people, including Dean Ball, former White House artificial intelligence adviser and current strategic future director of OpenAI, advocate that the United States should restrict or ban the use of China’s open source weighting model in disguise to maintain the United States’ technological dominance and prevent potential national security risks. via c