China weighs curbs on overseas access to advanced AI models

BEIJING: Chinese authorities have held discussions with leading technology companies over possible limits on overseas access to the country’s most advanced artificial intelligence models, including products that have not yet been released, according to three people familiar with the matter.

The talks, held over the past month, point to a broader effort by Beijing to keep domestically developed AI capabilities inside China. The discussions also reflect how China, like the United States, is increasingly treating cutting-edge AI as a strategic national asset subject to tighter controls.

Those attending the meetings included Alibaba, ByteDance and startup Z.ai, the people said. They were not authorised to speak publicly and declined to be named. Since the arrival of DeepSeek’s R1 model last year, Chinese AI systems have expanded their global reach on the back of lower costs and improving performance. Any move to narrow overseas access could have wider effects on AI markets, with costs for many businesses likely to rise.

Restrictions still under discussion

Two of the sources said the meetings were led by China’s Ministry of Commerce, with officials also discussing possible restrictions on both closed-source models and more open versions. One source said officials also discussed making the leaking or theft of proprietary AI technology an offence under China’s national security law. The same source added that officials raised the possibility of new steps to limit who can finance domestic AI startups.

Two sources said the exact scope of the proposed curbs remains under discussion and that the measures may apply only to future models. It is not yet clear when, or whether, the restrictions would be implemented.

China’s commerce ministry and the National Development and Reform Commission, whose officials also attended the meetings, did not respond to Reuters requests for comment. Alibaba, ByteDance and Z.ai also did not respond to Reuters queries. The three companies offer a mix of closed-source and open-weight AI systems that can be downloaded, run and modified by users. Alibaba’s Qwen and ByteDance’s Doubao are among the most widely used AI models in China, while Z.ai’s GLM-5.2 has recently attracted attention in Silicon Valley because its capabilities are seen as nearing leading US offerings at a much lower cost.

Security concerns and US comparison

Washington has also been focused on the national security implications of AI, particularly the risk that advanced American systems could be misused by military intelligence services in China, Russia and other countries of concern. In June, the administration of US President Donald Trump ordered that foreign nationals be denied access to Anthropic’s most advanced Fable and Mythos models.

That decision led Anthropic to disable the models for all users worldwide because nationality could not be verified in real time. Export controls on Fable, which is intended for the general public, were later lifted after additional safeguards were introduced. Mythos, which is designed for cybersecurity professionals, remains available only to some trusted US organisations.

Some US AI specialists have also argued that the United States should regulate the use of Chinese AI models. At the same time, two of the sources said Chinese authorities are particularly concerned that Mythos could be used to identify software vulnerabilities and that Washington could deploy the model against Chinese interests. Those concerns mirror public comments from state media and from Zhou Hongyi, founder of cybersecurity firm 360, who has said China should develop its own Mythos.

Broader Chinese measures on AI

China has introduced several steps this year aimed at protecting its domestic AI sector. In April, the state planner ordered Meta to unwind its $2 billion acquisition of Chinese-founded AI startup Manus. In early June, authorities issued broad new rules tightening oversight of overseas transactions involving Chinese investors, technology, data and national security.

According to two of the sources and a third person, China has also opened investigations this year into Manus and other local AI startups that moved abroad to determine whether they may have breached export control laws. Manus did not respond to requests for comment.

Reuters said it could not determine how any new restrictions on overseas access to Chinese AI models would work in practice. However, a May roundtable of Chinese legal experts on open-source AI rules offered possible clues. A summary published in an official journal of the Supreme People’s Court said participants proposed a tiered framework under which basic open-source tools would face a simple filing process, more advanced technologies would undergo security reviews, and the most sensitive frontier models would either be blocked from public release or limited to domestic use.