China's Ministry of Commerce and Ministry of Science and Technology have proposed adding 40 technologies to the country's export restriction catalogue, 23 of them in artificial intelligence. The announcement, No. 48 of 2026, was issued October 16. The catalogue changes are still a draft open for public comment until November 15, 2026. A separate registration requirement, however, is already in effect.
What AI technologies does China's export control draft cover?
The draft reaches well beyond chatbots. It covers frontier model weights, agent architectures and AI-enabled offensive cyber capability, and it sets numerical limits for each. Model weights, key checkpoints, source code, architecture and training and inference know-how are restricted if a model meets any one of three tests. The first is at least 10^26 FLOPs of training compute. The second is at least 2 trillion parameters. The third is a global top-10 position on public evaluation rankings, global call volume or comparable measures at the time of export.
Agents are treated separately from the models that power them. Planning, multi-agent coordination, tool use and game-theoretic decision technologies are restricted once they pass set benchmarks. High-performance agents built for critical infrastructure, defense or core government use would be prohibited outright. The draft's annex also includes AI malicious code generation and adaptive evasion, which would be prohibited, and AI-assisted vulnerability discovery, which would be restricted.
Does using a Chinese AI model through an API count as export?
Most of the commercial exposure depends on whether a company is sending a model out of China or simply letting customers use it. MOFCOM says API, app, web, SaaS and lightweight on-device use do not count as exporting a large model. Deploying a restricted model on infrastructure outside China does count, and a license would be required. Automatic licensing may apply when the deployment is on a Chinese company's own overseas cloud or compute center.
Does China's export registration requirement apply now?
The registration rule is the part that applies today. Until the revised catalogue takes effect, exporters must register through MOFCOM's unified business platform before shipping any listed technology abroad. Registration is triggered by four things: transferring technology ownership overseas, licensing domestic technology for overseas use, a change in ownership or control that gives a foreign entity control, and other cases MOFCOM designates. The registration form asks for recipient details, destination, a technology description, end use and estimated value. Exporters must decide for themselves whether a given technology falls within scope.
That structure creates risk for multinationals that never sell a product in China. Licensing and ownership changes can trigger registration, not just third-party sales. Analysts at ChinaTalk have read the form's questions on exports made outside formal contracts as potentially covering code shared within a corporate group, though that reading is an interpretation of the form rather than an official statement.
MOFCOM's spokesperson described the revision as routine and grounded in Article 20 of the Foreign Trade Law. The spokesperson said the revision "does not target any specific country, region, or entity," and that its aim is to prevent uncontrolled diffusion of technology while protecting legitimate cooperation.
Beyond AI, the announcement proposes 14 prohibited items and 26 restricted items in total. Five of the restricted items are in biomedicine and 12 are in photovoltaics. Under China's technology-trade framework, prohibited technologies cannot be exported at all, while restricted technologies require a license from commerce and science authorities. ChinaTalk notes that provincial authorities administer licensing, although central authorities have stepped in on prominent cases. The draft also excludes technology that is already public through open-source release or academic publication.
China has a track record here. In January 2026, MOFCOM announced an assessment of Meta's acquisition of Manus, a Singapore-registered AI agent company whose products were developed in China. The review examined whether moving staff and technology required export authorization. Companies with China-origin development histories are now likely to see similar questions in deal reviews.
The draft is not final, and its wording leaves room for debate. Analysts have pointed out that the "global leading" test names no ranking body, which means a model's status could change after it is deployed. The Cloud Security Alliance's AI Safety Initiative, in its analysis of the draft, recommended that organizations first inventory every Chinese-origin model, agent framework and AI security tool they use, and note whether each is consumed as a service or hosted locally. Locally hosted assets are the priority, the group said, because they fall within export management under the draft.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.