
Chinese entities are using advanced methods to leverage output from top American AI models to train their own domestic rivals, according to recent documents. This strategy, known as distillation, aims to strengthen China’s AI capabilities and counter the impact of tightening US restrictions on chip access.
A review of over 300 Chinese academic papers, patents, corporate disclosures, recruitment notices, and government documents reveals that distillation has moved beyond academic experiments to become a mainstream industrial practice in China. Chinese researchers utilize US models to generate answers, create computer code, and solve complex problems. Chinese companies then develop software to automate this process, and specialized workers prepare the material to train smaller domestic models.
Beijing’s push for cheaper AI deployment
The Chinese government is actively promoting distillation to make AI more affordable and accessible across its powerful manufacturing sector. While distillation is not illegal and is practiced in other parts of the world, it can help developers preserve specific skills in smaller, less expensive systems, potentially reducing their dependence on American technology.
China’s increasing use of distillation may undermine the effectiveness of US restrictions on advanced chip sales to China. By drawing on the capabilities of American AI models, Chinese developers can build their own AI systems, even as Washington imposes further limitations on China’s access to chips from companies like Nvidia.
Distillation as a point of contention
Distillation has become a source of tension between the world’s two largest economies, particularly ahead of the Asia-Pacific Economic Cooperation summit in Shenzhen, where US President Donald Trump and Chinese President Xi Jinping are expected to meet. In September, US authorities publicly accused Chinese companies of large-scale, targeted distillation of American AI models.
Related Post: Rubio’s America First visa push as U.S.-India ties deepen
During meetings in Chengdu in July, US Commerce Department officials expressed concerns to Chinese regulators about developers collecting output from leading American AI models. They warned that some activities could constitute intellectual property theft, according to three individuals familiar with the discussions.
Chinese officials defended distillation as a standard AI development technique, stating their willingness to take action in cases of IP theft but making no specific commitments, as reported by a US official familiar with the exchange.
From American answers to Chinese models
Distillation involves a powerful ‘teacher’ model generating answers or demonstrating task completion, which developers use to train a smaller ‘student’ system. For Chinese developers with limited computing power, this approach allows a single expensive query to perform repeated work.
An American model can provide the answer to a challenging question, enabling a Chinese team to retain valuable examples and train a local system around them. In a November 2024 paper, researchers from Shanghai Jiao Tong University described selecting 5,000 answers from OpenAI’s o1 model, translating them using GPT-4o mini, and assembling 10,750 bilingual training examples to further train Alibaba’s Qwen2.5-72B model.
In an April 2025 paper, Alibaba researchers outlined a more systematic approach, using OpenAI’s GPT-4 and GPT-4o to generate training material in non-Chinese languages. They then refined and scored responses before training smaller Qwen models.
Read Also: Money Expo México set for February 17-18
A February 2026 paper from researchers at Shanghai Jiao Tong University and Chinese food delivery company Meituan described using GPT-4o to generate plans for coordinating AI agents working on programming problems. This approach helped train a smaller model to manage such tasks.
Legal and geopolitical implications
Among leading American AI companies, OpenAI’s terms restrict using its output to develop competing models, subject to exceptions.
A US official familiar with the July Chengdu regulatory discussions stated that Washington believes industrial-scale distillation is the primary reason Chinese AI models have closed the gap with their American counterparts. However, two individuals familiar with Chinese developers’ operations disputed this, citing distillation as just one factor among advances in engineering, algorithms, data, and model design.
Chinese government documents do not explicitly instruct companies to distill American models but encourage distillation as part of Beijing’s initiative to expand AI across the world’s second-largest economy. A national AI+Manufacturing policy package, published by eight Chinese government agencies in January, includes guidance supporting model training, fine-tuning, and distillation for industrial applications.
Due to growing geopolitical concerns surrounding distillation, Chinese AI developers have recently sought guidance from Beijing’s industry regulator on the boundaries between ordinary use of foreign models and organized collection of output for training, according to an individual familiar with these discussions. In conclusion, the increasing use of distillation in China highlights the complexities of the US-China AI rivalry and the need for clear guidelines on the use of foreign AI models.