China tightens export controls on open-weight models and 23 key AI technologies
decoder ·
China's commerce and science ministries have released a draft rule tightening export controls on 23 key AI technologies, including frontier models, agents, and autonomous malware tools. Large language models face restrictions if they were trained on sensitive Chinese data or meet specific technical thresholds, such as having at least 2 trillion parameters, 10^26 FLOPs of training compute, or ranking in the global top 10. While these measures aim to restrict foreign access to critical capabilities, the plan exempts technology that is already publicly available through open-source releases or research papers. To support its domestic open-source ecosystem, China introduces a fast-track process granting automatic licensing for registered open-weight models. However, installing restricted AI tech on local servers abroad still requires standard export licenses, whereas API usage remains unrestricted. This regulatory shift may force US companies to rely more on domestic models and complicate Europe's strategy for maintaining AI sovereignty via open-weight deployments. Beijing also pledges to support AI development in the Global South through the World Artificial Intelligence Cooperation Organization.
decoder