Beijing accuses US firms of training AI models on Chinese examples

Beijing has accused US artificial intelligence companies of utilizing Chinese examples to train their models, following Washington's threat of sanctions on Chinese firms for allegedly stealing American technology. The Chinese Commerce Ministry stated that many American AI enterprises have incorporated Chinese models during their research, development, and training processes through a method known as distillation.

This technique involves training an AI model to replicate the output of a larger, more powerful one, and is commonly employed in the industry. However, US officials argue that Chinese companies have scaled up this method to unlawfully replicate American systems. The Chinese ministry dismissed the US accusations as lacking factual basis and legal grounds, noting their application of double standards and representing a form of artificial intelligence hegemonism.

China vowed to take all necessary measures to protect its legitimate rights and interests. Last week, a senior White House official alleged that Chinese startup Moonshot AI had covertly copied Anthropic's most advanced model to create its Kimi K3. The release of the Moonshot model sparked a new debate over the AI rivalry between the US and China.

Treasury Secretary Scott Bessent recently threatened sanctions against China, and reports suggest the US may also consider banning or imposing restrictions on foreign-made open-source models, many of which are developed through distillation.