China has held talks about limiting overseas access to and the export of advanced artificial intelligence models, and the U.S. used its AI kill switch in June. In July, China’s Ministry of Commerce convened meetings with major AI firms Alibaba, ByteDance and the startup Z.ai to discuss potentially restricting who can use the country’s most advanced models, including some not yet released. Officials discussed possible export limits and restrictions on users, while the scope and timing of any measures remain unresolved.
Chinese authorities held discussions with AI companies about limiting overseas use and export of advanced AI models, and participants discussed putting limits on both closed-source models and open-weight ones. Officials raised making any unauthorized disclosure or theft of proprietary AI technology an offense under China’s national security law. Separately, participants floated new measures to restrict which investors can fund domestic AI startups.
The scope of any potential restrictions is still being debated, and some sources say the measures may apply only to future models rather than existing ones, while proposals discussed included limits on overseas use of the most advanced models. No timeline has been set for any of the proposed measures, and it is not certain that anything discussed will come into force. The debate also covered which specific models would be included and whether investor restrictions should apply.
A summary published in a Supreme People’s Court journal from a May roundtable of Chinese legal experts proposed a three-tier structure for regulating open-source AI. Under the proposed framework, basic open-source tools would require a simple government filing before release. More advanced technologies would face security reviews prior to release.
The most sensitive frontier models would be barred from public release or restricted to domestic use only under the proposal. The summary framed these three tiers as progressively stricter regulatory requirements tied to technological sophistication. Alibaba’s Qwen series was identified among the domestic products discussed and has built a large following on Hugging Face. ByteDance’s Doubao was named as one of the dominant AI products inside China, and Z.ai’s GLM-5.2 was noted for matching top American models on some benchmarks while offering API access at a fraction of the cost.
The May roundtable focused on open-source AI regulation and produced the summary published in the judicial journal. The proposal specifies filing, review, and restriction measures corresponding to each tier. The document presents these measures as regulatory options considered by legal experts.
The published summary records expert recommendations. The proposal reflects regulatory options under consideration.
Chinese officials expressed cybersecurity concerns after the U.S. restricted Anthropic’s Mythos in June, noting the model had been subject to U.S. controls. Officials were alarmed that Mythos could be reverse-engineered and used against Chinese systems. These concerns formed part of discussions with domestic AI firms about potential limits on overseas access.
Participants noted that restricting overseas access to advanced Chinese models would likely raise costs for businesses that have relied on Chinese models as cheaper, less restricted alternatives to U.S. frontier systems. The potential economic impact was discussed alongside technical and legal measures under consideration. Proposals considered included making unauthorized disclosure or theft of proprietary AI technology an offense under China’s national security law.
Officials and industry participants also raised measures to limit which investors can fund domestic AI startups as part of the regulatory debate. The discussions took place amid talks convened by the Ministry of Commerce with major firms including Alibaba, ByteDance and Z.ai. The scope and timing of any changes remain under debate and no timeline or certainty of implementation has been established.
