China's Ministry of Commerce and Ministry of Science and Technology published a draft on October 16 adding 23 AI technologies to the catalogue of technologies prohibited or restricted from export, part of a wider 40-item revision. The AI list covers large models, agents, compute clusters, chip designs and data-processing tools, with 14 technologies proposed for a ban and 26 for restriction across the full catalogue.

For large models, core technology — weights, checkpoints, source code and architecture — would be controlled once a model reaches about 2 trillion parameters, 10^26 FLOPs of training compute, or a global top-10 ranking, and exports of core technology for models trained on sensitive national data would be banned. The draft names no models, though reporting notes Kimi K3 and Qwen 3.8-Max sit above the size threshold.

Access through APIs, apps, websites or SaaS would not count as an export, and technology already public through open source or academic papers sits outside the scope, with open-sourcing eligible for automatic licensing. Public comment runs until November 15, 2026, and an interim pre-export registration requirement is already in force.