US Treasury Secretary Scott Bessent warned on Tuesday that the United States will examine Chinese open-weight AI models for stolen intellectual property and could sanction their developers within weeks. On the same day, Nvidia chief executive Jensen Huang offered the opposite message, telling reporters that strong open-weight models from China should be used and run by American firms.<\/p>\n
Speaking on television, Bessent framed the issue as one of national IP policy rather than model regulation. “This administration supports open source models, but what we do not support is IP theft,” he said. His stated evidence: watermarks traced to US large language models appearing inside Chinese counterparts. “We are finding watermarks of our US large language models on many Chinese models,” Bessent said. He described the practice, technically known as distillation, as theft and promised US action within weeks. Bloomberg first reported the administration’s stance.<\/p>\n
The trigger was Kimi K3, an open-weight model released by China’s Moonshot AI. The model matches or beats leading systems from OpenAI and Anthropic on certain benchmarks while being far cheaper to run. Its release contributed to a chip-stock selloff and a fresh round of concern in Washington. US labs have accused Chinese competitors of siphoning their models for months, and the Kimi K3 debut sharpened the dispute.<\/p>\n
At a new chip plant in Texas, Huang told Axios that the panic is misplaced. “These Chinese models are excellent,” he said. “Open-source models that are excellent should be used.” He argued that American firms should “absolutely” be allowed to run them. His reasoning is commercial: cheaper, free models spread AI into more hands, and broader adoption drives demand for the chips, data centres, and power that Nvidia sells. “Free AI should be great for chips,” he said.<\/p>\n
Huang also pushed back on the idea that a downloaded Chinese model functions as a “backdoor” to Beijing, saying companies can isolate them in secure sandboxes.<\/p>\n
Distillation is widespread, and US labs practise it too. “We know distillation to be a very small factor,” Hugging Face chief Clem Delangue told TechCrunch. “It is a practice that everyone is doing, including companies in the US,” he added.<\/p>\n
There is also a pointed irony. Anthropic, one of the most vocal accusers of Chinese distillation, recently had a $1.5 billion settlement approved over pirated books used to train Claude. Huang went further, urging the government to stop restricting Anthropic’s own Mythos model. “Let Anthropic run,” he said.<\/p>\n
Bessent will lead the US side at AI talks with China scheduled for September, and the distillation dispute is expected to be on the agenda. For now the divide is sharp: the US government wants to fence Chinese models out, while the world’s most valuable chipmaker wants them let in.<\/p>\n
Bessent said the US will scrutinise Chinese open-weight models for stolen intellectual property and could sanction their developers. He cited watermarks from US large language models appearing in Chinese models and described the practice as theft, promising US action within weeks.<\/p>\n
Speaking at a new chip plant in Texas, Huang said strong open-weight models from China should be used and that American firms should be allowed to run them. He argued that free, widely used AI increases demand for Nvidia’s chips, data centres, and power.<\/p>\n
Distillation is the practice of training one model on the outputs of another. Bessent called it theft when US model watermarks appear in Chinese systems. Hugging Face’s Clem Delangue has said distillation is common and is practised by US companies as well.<\/p>\n