China plans to tighten export controls on open-weight models and 23 key AI technologies

China is moving to tighten export controls over 23 key AI technologies, covering model weights, source code, and training methods. The proposed rules target systems that reach 10^26 FLOPs or 2 trillion parameters, setting a clear threshold for what falls under the new restrictions.

A fast-track provision will still allow open-source releases to be cleared as soon as they are registered, signaling that Beijing intends to preserve some momentum for open-weight model distribution while adding oversight. This distinction between proprietary and open-source pathways could shape how Chinese labs release and share their models internationally.

The move is notable because open-weight models have been a central part of China's AI strategy, enabling rapid global adoption and community-driven development. Tightening controls on model weights and training methods could slow the flow of Chinese AI technology abroad and raise questions about how international teams access and build on Chinese models.

Analysts will watch whether the 10^26 FLOPs and 2 trillion parameter thresholds effectively capture frontier Chinese models or leave room for smaller but still capable systems to move freely. The registration-based fast track for open-source releases may serve as a compromise, balancing openness with regulatory control.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.