美国要动真格了：查“蒸馏”AI摸走对手IP？浮水印都给你揪出来了

U.S. Treasury Secretary Scott Bessent just signaled a tougher compliance era for China’s open-source AI models. The Trump administration says it will scrutinize whether Chinese model developers are engaging in “model distillation” without authorization—potentially using this to extract U.S. vendors’ capabilities and intellectual property. Bessent claimed the U.S. has already found “watermarks” in multiple Chinese AI models, allegedly tracing back to U.S. large language models. Over the next days to weeks, U.S. investigators plan to verify whether these models involved licensed/unauthorized distillation. He didn’t name the specific Chinese models, the alleged source models, or exactly how technical evidence would translate into an IP-violation determination. But he warned sanctions are on the table if infringement is confirmed. Beyond enforcement, there’s also a possible policy shift: U.S.-based companies using Chinese AI models may be required to disclose details to end users—though no specifics yet. Distillation is common, but the risk is “scraping” massive outputs via fake accounts and retraining competitors—exactly the kind of behavior regulators now appear to be hunting. #AIRegulation #IPLaw #ModelDistillation #OpenSourceAI #USChinaTech #MachineLearning

Want to learn more? Visit Explore the world, stay updated on travel insights and international affairs, and discover authentic stories from real life