China tightens export controls on open-weight models and 23 AI technologies
China's commerce and science ministries have issued a draft rule that expands the list of AI technologies subject to export controls to 23 items, covering large models, AI agents, compute clusters and data‑processing tools. The regulation targets models trained on data deemed "important" under China's Data Security Law, as well as any model that reaches at least 2 trillion parameters, 10^26…
Key points
- China’s commerce and science ministries released a draft rule restricting export of 23 AI technologies, including frontier models and AI agents.
- Models trained on "important data" or exceeding 2 trillion parameters, 10^26 FLOPs, or a global top‑10 ranking face export controls.
- Open‑source releases and publicly available AI tools are exempt, and a fast‑track process grants automatic licensing for qualifying exports.
Open‑source releases and AI tools already public through research papers are completely exempt, and a fast‑track process grants automatic licensing for qualifying open‑weight exports. Companies can continue to use restricted models via API calls, SaaS tools or lightweight local versions without a license, but installing the technology on foreign servers requires a standard export permit. Beijing also says it will support AI development in the Global South and ease exports for capacity‑building projects under the World Artificial Intelligence Cooperation Organization.
Analysts note the rules could push U.S. firms that rely on open‑weight AI to double down on domestic models, while Europe faces a tougher environment for maintaining AI sovereignty.
China tightens export controls on open-weight models and 23 key AI technologies
The Decoder · 16 October 2026
Loading the full article…
This text was published by The Decoder and written by Matthias Bastian. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Policy & Regulation
All →- OpenAI warns courts may lack reliable deepfake detection methods · 1 src
- White House AI task force starts 120-day review of superintelligence risks · 2 src
- OpenAI, Anthropic, Meta models breach external systems, raising liability questions · 2 src
- AI espionage is a reality, says DNI Jay Clayton · 1 src
- Nvidia CEO Jensen Huang says U.S. must work with Chinese AI companies on safety · 6 src
Comments
via GitHub Discussions