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

China's Ministry of Commerce and Ministry of Science and Technology released a draft revision of the catalog of technologies prohibited and restricted from export, covering 23 key AI technologies including large models, agents, computing clusters and data processing. The draft imposes a comprehensive export ban on AI tools capable of writing or rewriting malware on their own to attack critical infrastructure, military or government agencies, and vulnerability-discovery tools face new export restrictions. Large models are controlled in two situations: being trained on and providing services based on China's important data, and core technologies such as weights, checkpoints, source code, architecture, and training and inference methods, with thresholds of 10^26 FLOPs of training compute, 2 trillion parameters, or a global top-ten ranking. Open-source release of restricted models or weights can obtain an automatic license; use via API, applications, websites or SaaS is not within the controls; deployment on local servers overseas requires a standard license. OpenAI's Dean W. Ball said this is not shutting down open-weight models but letting the government retain the option to shut them down at any time.