China drafts tighter export controls on open-weight models and 23 AI technologies
China's commerce and science ministries released a draft update to the national export control catalogue covering 23 key AI technologies, including large models, AI agents, compute clusters and data processing. Tools that can autonomously write or alter malware to attack critical infrastructure, the military or government agencies face a total export ban, while AI vulnerability-discovery tools get new export limits. Large-model controls apply when a model is trained on China's 'important data' and serves users based on it, or when core technologies such as weights, checkpoints, source code, architecture and training methods are involved - triggering at 10^26 FLOPs of training compute, at least 2 trillion parameters, or a global top-10 ranking. To protect its open-source boom, publishing restricted tech and weights as open source qualifies for automatic licensing; API, app, website and SaaS use stays exempt, but installing restricted AI tech on local servers abroad needs a standard license. OpenAI's Dean W. Ball said the plan does not shut down open-weight models but gives Beijing the option to do so.
Beijing would gain the option to restrict future open-weight model exports, which could push US and European firms that rely on Chinese open models toward domestic alternatives.
China Ministry of Commerce (Mofcom)China Ministry of Science and TechnologyOpenAI