Bessent says U.S. could sanction China over AI model 'theft' - CNBC

What the report says

CNBC reported that U.S. Treasury Secretary Scott Bessent said Tuesday the Trump administration plans to examine whether Chinese artificial intelligence models have been built by distilling technology from U.S. systems. In an interview on Fox Business, Bessent said Washington could use sanctions if overseas models are found to have taken intellectual property from American companies, framing the issue as one of “IP theft.”

The remarks come as Chinese open-weight AI models draw increasing attention in competition with products from U.S. companies such as OpenAI and Anthropic. CNBC cited Moonshot AI’s recently released Kimi K3 model as one example, saying it has surpassed those companies on some industry benchmarks. Open-weight models make trained parameters available publicly, though their code and training data are not necessarily disclosed.

Bessent referred to distillation, a training technique in which a smaller model is developed using outputs from a more capable model. CNBC noted that Anthropic recently wrote to the U.S. Senate Banking Committee alleging Alibaba had carried out a major distillation attack against it. Bessent also said U.S. officials are finding watermarks from American large language models in Chinese models and that the administration would review the issue in the coming days or weeks.

The dispute adds to broader U.S.-China tensions over AI leadership, technology controls and intellectual property. CNBC also noted that Reuters reported U.S.-China AI talks are expected in September, with Bessent representing the U.S. The article added context that OpenAI and Anthropic have themselves faced copyright-related claims over training data.