Technology news and discussion on TalkFuse.
China Caught Copying American AI’s Homework, Hundreds Of Documents Show
A Reuters review of more than 300 Chinese academic papers, government files and other records found that Chinese researchers and companies have used outputs from leading US AI models to train their own systems. This practice, known as distillation, is common worldwide, but may help Chinese firms develop capable models with fewer restricted chips, raising questions about intellectual property and the effectiveness of US export controls.
Examples include research using OpenAI’s GPT-4 and GPT-4o outputs to train Alibaba’s Qwen models, and a paper describing training an Alibaba model on OpenAI’s o1 outputs. China’s embassy called US allegations unfounded and described distillation as normal commercial practice; the US Commerce Department has warned that some cases may amount to intellectual property theft. Alibaba’s Qwen4-Max ranked sixth in Code Arena, while experts also raised concerns about supply-chain risks and dependence on Chinese AI.
In short
- Chinese research papers document training models on US AI outputs.
- Distillation may help firms develop AI with fewer restricted chips.
- China disputes US claims; experts cite IP and supply-chain concerns.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Distilling outputs from an AI model can be a legitimate way to learn from publicly accessible information, and researchers may see it as ordinary model development rather than copying protected code or weights. Chinese firms could argue that such methods widen access to AI capabilities and that US export controls make efficient training especially important; the reported use of US model outputs also does not by itself establish a legal violation.
The case against
Even when distillation is a common technique, systematically using a competitor’s outputs to train a commercial rival may appropriate the value of costly research and raise unresolved intellectual-property questions. US officials and companies could argue that this practice helps Chinese firms get around chip restrictions, weakens the intended effect of export controls, and creates strategic or supply-chain risks if critical systems become dependent on models whose development and oversight are less transparent.
AI Americas Asia Politics Technology World