China AI Export Controls Could Lock Down Top Models

China AI export controls are suddenly the biggest question hanging over the free Chinese models many developers rely on. Beijing has proposed restricting or banning exports of forty technologies spanning artificial intelligence, chips, solar power and biomedicine, and twenty-three of them are AI.

What China actually proposed

According to PBS NewsHour, the Ministry of Commerce and the Ministry of Science and Technology jointly published a draft update to China's catalog of technologies subject to export bans or restrictions. A commerce ministry spokesperson called the move necessary and cited national security as one reason.

The AI items range from advanced large models to agents. Exports of core technology for models trained on sensitive national data would be prohibited, along with agents built for highly sensitive settings such as national defence and critical information infrastructure.

The two trillion parameter line

Here is the part that matters for the AI race. Exports of models with parameter counts of two trillion or more, or with certain world-leading capabilities, could be restricted. Parameter count is a rough measure of how much a model can learn.

PBS notes that Chinese models including Moonshot's Kimi K3 and Alibaba's Qwen 3.8-Max exceed two trillion parameters, although the proposal did not name any specific model. The draft also lists some chip design technologies, including for certain AI processors and memory chips.

Why open access is in the spotlight

A New York Times news analysis argues that Beijing is moving to tighten its grip on the global spread of its top-tier AI. As the Times tells it, Chinese companies such as Moonshot, DeepSeek and Z.ai turned free downloads into a major selling point, and developers have come to see them as an alternative to services from American labs such as Anthropic and OpenAI. You can read more at The New York Times.

The analysis adds that the draft rules threw uncertainty into that market. The proposal would also ban the export of AI that can be used to launch cyberattacks, plus models trained on what China calls core data.

What stays free

Not everything is locked down. Reports on the draft say calls to AI through application programming interfaces, websites, mobile apps and software services would remain outside the proposed restrictions. That distinction separates handing over the underlying technology from simply letting people use a service.

The ministries also say technology already in the public domain, through open source or published academic papers, would not fall under the controls. Public comment on the draft runs to November fifteenth, and the technologies are subject to temporary registration before export in the meantime.

The bigger geopolitical picture

The timing is odd. PBS reports the proposal arrived as China and the United States made progress on AI safety dialogue and on a notification channel for AI incidents that could threaten national security. At the same time, U.S.-led restrictions have long blocked China from the most advanced AI chips and chipmaking machines.

Omdia analyst Lian Jye Su told PBS the move can be seen as a response to Western countries setting sovereignty requirements on AI, robotics and telecoms. Natixis economist Gary Ng said the world is getting more fragmented in trade and investment, and also in access to technology.

Reading between the lines

Notice what the draft does not do. It does not ban Chinese AI apps, and it does not stop people abroad from chatting with a Chinese model online. It targets the crown jewels: model weights, source code, design and training know-how, and the most capable agents.

That fits the stated goal. The commerce ministry said the measures aim at reinforcing the security bottom line, preventing misuse and abuse, and ensuring that AI remains under human control. Critics will ask whether that is security policy or a way to keep a technological edge at home, and the honest answer is probably both.

It is also worth remembering this is still a draft. Drafts get softened, delayed or reshaped after public comment, so nobody can say yet how many of the models people use today would actually be caught.

What it means for builders and users

If you use a Chinese chatbot through its app, little changes for now. If you download open weights to run at home or build a startup on them, watch the final rules closely, because the draft could change what is allowed to leave China. For more, see our tech coverage.

The big takeaway is that AI is now treated like a strategic resource, the way chips and oil have been. China AI export controls would make the next frontier model from Beijing harder to download, and the final text will show how tough Beijing really wants to be.