China's First Move to Wall Off Its Own Frontier AI — Qwen, Doubao, GLM-5.2 and DeepSeek R1 All on the List
TL;DR
Reuters exclusive July 7: China's MOFCOM convened Alibaba, ByteDance and Z.ai to discuss tiered export curbs on top domestic AI models — Qwen, Doubao, GLM-5.2 and DeepSeek R1 all named.
Reuters reported on July 7, 2026 that over the past month China's Ministry of Commerce has convened Alibaba, ByteDance and Z.ai (Zhipu) to discuss limiting overseas access to China's top domestic AI models. It's the first time Beijing has flipped the export-control logic it has spent years fighting on the U.S. chip side and pointed it at its own open-weight LLMs.
The models named in the meetings include Alibaba's Qwen, ByteDance's Doubao, Z.ai's GLM-5.2 and DeepSeek's R1. Officials from the National Development and Reform Commission also attended.
Officials proposed a tiered regime: basic open-source tools would require a simple filing; more capable models would face security reviews; the most sensitive frontier models would be barred from public release or restricted to domestic use only. The meetings also discussed making the leak or theft of proprietary AI technology a prosecutable offence under China's national security law, and floated new limits on who can fund domestic AI startups.
Scope is still contested. Three sources told Reuters the restrictions may only apply to future models, and it is not yet clear when or even if they take effect. MOFCOM, the NDRC, Alibaba, ByteDance and Z.ai did not respond to requests for comment.
Six months ago DeepSeek R1 turned Chinese open-source models into the default choice for global developers at a fraction of the cost of U.S. closed-source rivals. That openness is why Qwen has held the top of Hugging Face for months. If Beijing shuts the door, inference costs for anyone building on Chinese open weights snap upward overnight.
If it works, Beijing turns "cheap and good" into a bargaining chip and pushes Washington to soften chip controls. If it doesn't, the global open-source share Qwen vacates gets filled inside six months by Meta's Llama and the UAE's Falcon.
via Reuters / Yahoo Finance / AOL
The models named in the meetings include Alibaba's Qwen, ByteDance's Doubao, Z.ai's GLM-5.2 and DeepSeek's R1. Officials from the National Development and Reform Commission also attended.
Officials proposed a tiered regime: basic open-source tools would require a simple filing; more capable models would face security reviews; the most sensitive frontier models would be barred from public release or restricted to domestic use only. The meetings also discussed making the leak or theft of proprietary AI technology a prosecutable offence under China's national security law, and floated new limits on who can fund domestic AI startups.
Scope is still contested. Three sources told Reuters the restrictions may only apply to future models, and it is not yet clear when or even if they take effect. MOFCOM, the NDRC, Alibaba, ByteDance and Z.ai did not respond to requests for comment.
Six months ago DeepSeek R1 turned Chinese open-source models into the default choice for global developers at a fraction of the cost of U.S. closed-source rivals. That openness is why Qwen has held the top of Hugging Face for months. If Beijing shuts the door, inference costs for anyone building on Chinese open weights snap upward overnight.
If it works, Beijing turns "cheap and good" into a bargaining chip and pushes Washington to soften chip controls. If it doesn't, the global open-source share Qwen vacates gets filled inside six months by Meta's Llama and the UAE's Falcon.
via Reuters / Yahoo Finance / AOL
