Chinese labs use OpenAI model outputs to train domestic rivals, Reuters review finds

Nov 17, 2026
Papers from Shanghai Jiao Tong, Alibaba and Meituan describe distilling o1 and GPT-4o answers into smaller Chinese models.

Chinese companies and researchers increasingly use output from leading American AI models to train domestic rivals, according to a Reuters review of more than 300 academic papers, patents, corporate disclosures, recruitment notices and government documents, AI Weekly and technology.org report. The technique, known as distillation, uses a strong model's answers to teach a smaller one.

In a November 2024 study, Shanghai Jiao Tong University researchers turned about 5,000 answers from OpenAI's o1 model into 10,750 bilingual training examples for Alibaba's Qwen2.5-72B. An April 2025 Alibaba paper describes using GPT-4 and GPT-4o to generate training material for smaller Qwen models, and a February 2026 paper by Shanghai Jiao Tong and Meituan used GPT-4o to draft plans for coordinating AI agents.

A US official said industrial-scale distillation is the main reason Chinese models have narrowed the gap with American ones, and US Commerce Department officials told Chinese regulators in July that some collection of model output could amount to intellectual-property theft. Chinese government documents do not tell companies to distill American models but encourage distillation to spread AI across the economy. Distillation is not illegal in itself, but OpenAI and Anthropic restrict using their outputs to build competing systems.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Daily AI Brief.

Also, consider following us on social media:

Free newsletter

Daily AI Brief

Daily report covering major AI developments and industry news, with both top stories and complete market updates

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more
Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.