China's commerce and science ministries are moving to tighten artificial intelligence export controls by releasing a draft rule that targets 23 key areas, including frontier models, AI agents, and autonomous malware tools. The proposed catalog introduces strict limits on core technologies based on training compute, parameter size, or global rankings.
Model restrictions will trigger if a system was trained on sensitive Chinese data, or if it crosses specific technical thresholds such as 2 trillion parameters, 10^26 FLOPs of training compute, or a global top-10 ranking. To prevent hindering domestic AI development, the plan establishes a fast-track process for open-weight releases, granting automatic approval upon registration while already public technology remains entirely exempt.
This regulatory framework mirrors export controls implemented by the United States and gives Beijing the legal mechanisms to regulate powerful domestic models. The policy impacts international strategy by encouraging US companies to rely on domestic alternatives while complicating technological sovereignty efforts in Europe.