Beijing Weighs Curbing Overseas Access to Top Chinese AI Models

Reuters reported that China's Ministry of Commerce has spent the past month sitting down with Alibaba, ByteDance and startup Z.ai to talk through what it would take to limit overseas access to the country's most advanced AI models, both closed-source and more open versions. That is a striking pivot. For most of the last eighteen months, Chinese labs have been shipping models globally as fast as possible, and cheap Chinese open weights have been a big part of why small teams outside the US could afford to run frontier-ish AI at all.

The specific targets are the models everyone actually uses: Alibaba's Qwen, ByteDance's Doubao, and Z.ai's GLM-5.2, which the report describes as coming close to leading US offerings at a fraction of the cost. Participants reportedly floated a tiered regime, with basic open-source tools subject to a simple filing, more advanced technologies facing security reviews, and the most sensitive frontier models either barred from public release or restricted to domestic use. Officials also discussed making any leak or theft of proprietary AI technology an offence under China's national security law, and raised the possibility of new measures to restrict who can fund domestic AI startups.

The reason this matters even if you never touch a Chinese model is that the presence of good, cheap open weights has been a real ceiling on what US labs could charge globally. A RAND report cited in the piece found Chinese LLMs' global market share rose to 13% from 3% in the two months after DeepSeek launched its R1 model in January last year, with the biggest gains in developing countries and in nations with close ties to Beijing. If Qwen and GLM effectively become domestic-only for the top tier, that pricing pressure lifts.

The honest caveat is that this is a Reuters exclusive built on three people familiar with the discussions who were not authorised to speak on the record, and the sources themselves say the measures may only apply to future models. There is no draft rule, no effective date, and no clarity on whether 'restrict overseas access' means geo-blocking APIs, halting future open-weight drops, or something narrower. Weights that have already shipped globally are, in practical terms, uncollectable.

Still, the direction of travel is the part worth watching. Both Washington and Beijing are now treating frontier model weights as a national asset rather than a shipping product, and the teams that benefited most from the Chinese open-weight wave should be modelling what their stack looks like if the next Qwen or GLM never leaves the mainland.