Chinese regulators have held recent talks with major tech firms about potentially restricting overseas access to the country’s most advanced artificial intelligence models, according to three people familiar with the discussions.
Those meetings, held over the past month, signal Beijing’s growing effort to treat leading AI capabilities as strategic assets deserving tighter controls, similar to measures pursued by other governments.
Participants in the discussions included Alibaba, ByteDance and the startup Z.ai, the sources said. The meetings were led by the Ministry of Commerce and attended by officials from the National Development and Reform Commission, though the precise scope and timing of any measures remain undecided.
Officials reportedly focused on limiting cross-border access to state-of-the-art models — covering both closed-source systems and certain open variants — and discussed stronger penalties for leaks or theft of proprietary AI technology, potentially classifying such acts under national security law. They also explored tighter rules on who may invest in domestic AI startups.
The conversation comes as Chinese models such as DeepSeek’s R1 and newer offerings have gained international users because of lower costs and improving capabilities. Any official decision to restrict foreign access could raise costs for businesses that rely on these models and reshape parts of the global AI market.
Among domestic models mentioned are Alibaba’s Qwen and ByteDance’s Doubao, while Z.ai’s GLM-5.2 has drawn attention in Silicon Valley for performance close to leading U.S. models at a lower price point. Companies contacted for comment did not respond, and Chinese agencies overseeing export and planning did not reply to requests for comment.
The deliberations are partly framed by international precedents. U.S. restrictions this year on access to advanced Anthropic models — which led the firm to disable some models globally because of nationality-verification limits — have heightened concerns about the cross-border risks of frontier AI. U.S. officials and some experts have warned about potential misuse of powerful models by foreign states.
Chinese sources said Beijing is particularly anxious about tools like Anthropic’s Mythos, which is designed for cybersecurity professionals, fearing such models could be used to probe or exploit software vulnerabilities against Chinese targets. Those concerns have been echoed in Chinese state media and by cybersecurity figures in China.
China has already taken several steps this year to tighten control over AI-related technology and investments. Reported actions include ordering the unwinding of a foreign acquisition tied to a Chinese-founded AI startup, issuing new rules tightening oversight of overseas deals that involve Chinese technology or investors, and investigating some local AI firms that moved operations abroad for possible export-control breaches.
How any new limits on overseas access to Chinese AI models would be implemented is not clear. A recent roundtable of legal experts, summarized in a Supreme People’s Court journal, proposed a tiered approach: basic open-source tools would face light filings, more advanced technologies would undergo security reviews, and the most sensitive frontier models could be barred from public release or restricted to domestic use.
Sources stressed the proposals under discussion may apply only to future models and that no formal decision has been announced. Reuters and others seeking comment were unable to confirm details or timing. If enacted, these measures would reflect a broader global trend toward treating cutting-edge AI as a matter of national security and economic policy, with significant implications for developers, investors and international users.