China’s proposed AI export controls could reshape access to its next models

China has opened a consultation on export controls that could affect advanced AI models, AI agents and chip-related technology. The proposals are not yet rules, but they raise a practical question for European users: might the next Chinese models be available only through services hosted in China?

Watch Desk analysis

What happened

China’s Ministry of Commerce began the consultation on 16 October, with comments due by 15 November, according to Bruegel. The proposed controls could cover models trained on sensitive Chinese data; agents used in critical infrastructure, defence and government; advanced processing and memory chip-design know-how; and technology for linking more than 10,000 AI chips in one system.

Bruegel says current public models including DeepSeek 4 and Xiaomi MiMo would not be covered if they were not trained on sensitive data. The analysis says the proposals would distinguish between downloading a model for use on local servers and accessing it through a Chinese website, app or online service. The former could be restricted for covered models, while online access would remain available.

Why it matters

That distinction could change what “open” access means in practice. A model available only through a provider’s servers may be harder to use where data-residency rules apply, or where organisations do not want commercial information processed abroad. Bruegel also argues that uncertainty over future downloads could benefit European open-model developers such as Mistral.

For now, these are proposed measures, not a settled export regime. Their scope and final form will matter as much as the headline: a rule aimed at the most advanced models could leave today’s options untouched while changing the prospects for what comes next.

Our read

This is a consequential policy signal, not a reason to declare Chinese models suddenly off limits. The useful question is whether the consultation leads to rules that make access to future models depend on where they are hosted, what data trained them and how much compute they require. That is a lot of fine print for a technology often described as “open”.

What to watch

• The consultation’s outcome after its 15 November close.
• Whether any final rules define sensitive data or advanced models more narrowly.
• Whether European developers gain ground if access to future Chinese models becomes less predictable.

Discussion spark: If access to future Chinese models depends on using services hosted in China, would you still regard them as meaningfully open?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.