China Drafts Tighter Export Controls on Frontier AI Models, Exempts Open Source via Fast Track

China's Ministry of Commerce and Ministry of Science and Technology released a draft update on October 16, proposing to add 23 key AI technologies to the national export control list. These include large models, agents, compute clusters, and data processing tools. AI tools capable of autonomously writing or altering malware to attack critical infrastructure face a total export ban.

Export controls on large AI models trigger in two cases: first, if the model is trained on China's "important data" and offers services based on it; second, if core technologies (weights, checkpoints, source code, architecture, training/inference methods) meet specific thresholds—10^26 FLOPs of training compute, at least 2 trillion parameters, or a global top-10 ranking determined by major evaluators.

To avoid stifling its domestic open-source boom, the draft introduces a fast-track process. Publishing restricted AI tech and weights as open source qualifies for "automatic licensing," cleared immediately upon filing paperwork. Any AI technology already public through open-source releases or research papers is completely exempt from these new rules. Using restricted models via API, SaaS, or local lightweight versions remains free of controls, though installing them on overseas local servers requires a standard license.

The rules are currently in draft form. Until final implementation, companies must register affected exports through the commerce ministry's portal. This move may push US companies relying on open-weight AI toward domestic alternatives, while complicating Europe's strategy of using open-weight models for cloud sovereignty.