Chinese authorities have discussed possible controls on the global availability of the country’s most advanced artificial intelligence systems, with officials considering measures that could limit overseas access to future AI models and strengthen penalties for technology leaks, according to people familiar with the discussions.
Meetings were held over the past month between government officials and leading Chinese technology companies, including Alibaba, ByteDance and AI startup Z.ai, as Beijing examines ways to protect domestic artificial intelligence capabilities. The discussions reflect China’s growing view that advanced AI technology is a strategic national asset requiring greater oversight and protection.
The talks were led by China’s Ministry of Commerce, with officials also from the National Development and Reform Commission attending, according to sources who were not authorised to speak publicly. The potential restrictions remain under discussion, and it is unclear whether they will be implemented or when any measures could take effect.
Two sources said officials discussed possible limits on the overseas availability of the most advanced AI models, including both closed-source systems and more openly available versions that allow users to download and modify the underlying technology.
Authorities also discussed making the leak or theft of proprietary artificial intelligence technology punishable under China’s national security laws, one source said. Another source said officials raised the possibility of introducing new restrictions on who can provide funding to domestic AI startups.
The discussions come as Chinese AI models have gained global attention following the emergence of DeepSeek’s R1 model last year. Chinese companies have increasingly competed with leading international AI developers by offering systems with lower operating costs and rapidly improving capabilities.
Any restrictions on access to Chinese AI models could have wider effects on global technology markets, potentially increasing costs for businesses that rely on these systems, according to analysts familiar with the industry.
Alibaba’s Qwen and ByteDance’s Doubao are among China’s most widely used AI models. Z.ai has also attracted international attention with its GLM-5.2 model, which sources said has demonstrated capabilities approaching leading US systems while being developed at a lower cost.
The possible restrictions form part of a broader effort by Beijing to maintain control over sensitive technologies. China has introduced several measures this year aimed at protecting domestic AI development, including tighter oversight of overseas transactions involving Chinese investors, technology, data and national security concerns.
Earlier this year, Chinese authorities also investigated some domestic AI startups that had expanded overseas, including Manus, to examine whether they had breached export control rules, according to sources.
The discussions in China come alongside similar concerns in the United States, where officials have also focused on the national security implications of advanced artificial intelligence technologies. The US government has introduced restrictions affecting access to certain advanced AI models, citing concerns that foreign actors could misuse the technology.
In June, the US administration restricted foreign access to Anthropic’s advanced Fable and Mythos models, although some restrictions on Fable were later lifted after additional safeguards were introduced. Mythos, designed for cybersecurity professionals, remains available only to selected trusted organisations.
Chinese officials have reportedly expressed particular concern about the potential misuse of advanced cybersecurity-related AI systems. Two sources said authorities are worried that such technology could be used to exploit software vulnerabilities or threaten Chinese interests.
Possible future restrictions in China may follow a tiered approach discussed by Chinese legal experts, according to a summary published in an official Supreme People’s Court journal. The proposals suggested that basic open-source AI tools could face simple filing requirements, while more advanced technologies could require security reviews, and highly sensitive frontier models could be restricted from public release or limited to domestic use.
