The Chinese Ministry of Commerce and the Ministry of Science and Technology published a draft proposal on October 16 to restrict the export of AI technologies. The rules target large language models, AI agents, and chip design technologies.
Specific restrictions apply to neural networks trained on critical Chinese data, systems with computing power of 10 to the power of 26 FLOP, and models with two trillion or more parameters. Technologies ranking in the global top ten are also covered.
Commercial services, including official APIs, mobile applications, and cloud platforms, remain exempt from these controls. However, the export of open-source models with accessible weights remains a point of regulatory uncertainty.
Legal experts suggest the proposal responds to the attempted acquisition of Manus by Meta. The Chinese government blocked the $2 billion deal in April 2026, citing national security concerns, despite Manus being registered in Singapore.
Public consultation on the draft regulations is scheduled to run until November 15. Analysts note that the final impact will depend on the specific mechanisms for approving the export of open-source models and the application of exemptions.
