MOFCOM Announcement No. 48 proposes China's first frontier AI export controls

GoKawiil's interpretation of the reporting above, not reported fact.

The thresholds—10^26 FLOPs, 2 trillion parameters, or top-10 global ranking—suggest Beijing is targeting only the most capable models and agent systems rather than AI broadly, which could reshape how Chinese firms license or transfer leading-edge models and components abroad. Because the registration rule is already binding while the catalogue itself remains a draft, companies face near-term compliance uncertainty even before the scope of controls is finalized. The move also marks an expansion from China's narrower 2020 controls on recommendation and interaction technologies toward governing frontier-scale AI infrastructure and autonomous agent capabilities directly.

• Announcement No. 48 proposes adding 40 technologies, including 23 AI-related ones, to China's export control catalogue.
• A pre-export registration requirement is already in effect, covering even exports already in progress, while the catalogue changes remain open for comment until November 15, 2026.
• Controls target frontier-scale model weights, agent architectures, and offensive cyber tools using specific compute, parameter, and performance-ranking thresholds.