China considers export curbs on advanced AI models over Mythos vulnerability fears
China's Ministry of Commerce met with Alibaba, ByteDance, and Z.ai over the past month to discuss restricting foreign access to top-tier Chinese models, including unreleased ones.
China's Ministry of Commerce has held meetings with major tech firms over the past month to discuss potential restrictions on foreign access to advanced Chinese AI models, including those not yet released, according to three sources familiar with the discussions.
The talks involved at least Alibaba, ByteDance, and Z.ai, the maker of GLM-5.2. The scope of potential restrictions remains under discussion and may apply only to future models. It's unclear when or whether the measures will take effect.
Chinese officials are particularly concerned about vulnerabilities discovered by Anthropic's Claude Mythos model and the possibility that Washington could leverage Mythos against Chinese interests. Two sources said the threat is being taken seriously at policy levels.
The deliberations reflect Beijing's tension between promoting open-weight AI development and protecting strategic capabilities. If implemented, the restrictions would mark a significant shift in China's approach to AI openness. Alibaba's Qwen series, ByteDance's Doubao models, and Z.ai's GLM family have been among the most widely adopted open Chinese models in the global practitioner community.
The timing of any policy remains uncertain. Sources suggested restrictions might not materialize in 2026 or even 2027, but that pressure will mount as capabilities advance. The Commerce Ministry has not publicly commented on the meetings.
