Amid the controversy of ex-Nvidia CEO Jensen Huang’s fall from grace, the Trump administration announced new rules on Monday intended to prevent China from remotely accessing embargoed Nvidia chips. But that loophole isn’t the only trick Beijing has up its sleeve for sidestepping U.S. export restrictions.
According to a Tuesday report from Reuters, developers in China are getting increasingly creative in their use of distillation, a common industry practice through which new AI models are trained on the outputs of larger “teacher” models. While not technically illegal in the U.S.—OpenAI and Anthropic use it to train new versions of their proprietary models—distillation has become an intensely fraught subject in recent months amid the high-stakes AI race between the U.S. and China.
Based on a review of hundreds of academic papers, patents, corporate notices, and other official documents, the Reuters report found that distillation has become a widespread practice not only within niche AI circles in China, but throughout Chinese industry writ large. Chinese government documents reviewed by Reuters reportedly encourage distillation as a means of advancing AI usage across the country—without explicitly promoting the distillation of models from American companies like OpenAI and Anthropic.
Both of those companies have publicly accused leading Chinese AI labs of large-scale, illicit distillation campaigns.



