U.S. Treasury Secretary Scott Bessent says the Trump administration is looking into whether Chinese AI developers used U.S. intellectual property to create or improve open AI models. In comments reported by multiple outlets, Bessent says the U.S. is “finding watermarks” associated with American large language models in Chinese AI systems. He frames the issue as IP theft rather than open-source development, adding that the administration supports open models but does not support stealing from U.S. companies. Several reports say the U.S. could consider sanctions if it finds evidence that overseas models were derived from or “distilled” from U.S. counterparts without authorization.
The warnings build on broader concerns previously raised by U.S. AI companies. At least one report notes that OpenAI and Anthropic sent letters to the U.S. government about alleged unauthorized use of their data and models by Chinese actors. The administration’s approach, as described in the reporting, is to examine overseas open-source AI systems carefully and take action if evidence meets the government’s threshold. Separately, one report cites upcoming U.S.-China AI discussions scheduled for September.
