China proposes export controls on 23 AI technologies: weights of 2T-parameter or global top-10 models restricted, open source and API services exempt

On Oct 16, 2026, China's Ministry of Commerce and Ministry of Science and Technology issued Announcement No. 48, proposing to add 40 technologies to the Catalogue of Technologies Prohibited or Restricted from Export, 23 of them in AI (large models, agents, compute clusters, information processing). Comments are open until Nov 15. Under the draft annex, restricted model technology covers models trained on "important data" and the core technology (weights, checkpoints, source code, architecture, training and inference acceleration) of models with any of: training compute of at least 10^26 FLOPs, at least 2 trillion parameters, or a global top-10 ranking on mainstream evaluations. Agents are restricted above concrete thresholds, such as decomposing instructions into 20 or more sub-tasks with at least 90% average success, 60 or more concurrent collaborating agents, or 200 or more integrated third-party tools. AI-driven malware generation and mutation is on the prohibited list, and AI-assisted vulnerability discovery (for example, 75% or higher on authoritative benchmarks) is restricted. The ministry says technology already open-sourced or published in papers is out of scope, and serving a restricted model abroad via API, app or web, SaaS, or lightweight on-device deployment is not a technology export; overseas on-premise deployment is, though deployment on a company's own overseas cloud can get automatic licensing. A temporary pre-export registration requirement took effect on Oct 16 and applies to transfers, licenses, and changes of control to foreign entities, including deals already underway.