For the past two years, China's answer to America's lead in artificial intelligence has been to give its technology away. Chinese companies released powerful "open-weight" AI models, ones whose underlying settings anyone can download and run for free, and those models spread rapidly around the world. Now Beijing may be reconsidering that openness. Chinese officials have discussed restricting overseas access to the country's most advanced AI models, Reuters reported, citing people familiar with the matter.
What is reportedly being discussed
According to the report, China's Ministry of Commerce, with the National Development and Reform Commission also involved, has met technology firms to weigh limits on the most capable AI systems, both the closed kind and the more open versions. Officials floated a tiered approach: basic open-source tools would need only a simple filing, more advanced technology would face security reviews, and the most sensitive "frontier" models could be barred from public release or restricted to use inside China, Reuters reported.
The companies said to be involved in the discussions include Alibaba, whose Qwen models are among the most downloaded in the world, ByteDance, maker of the Doubao model, and Z.ai, behind the GLM series. Officials also discussed making the theft of AI models an offence under China's national-security law and controlling who can fund domestic AI startups.
Two cautions are essential. Beijing has announced nothing, and neither ministry responded to Reuters' request for comment. The sources described the plans as still being worked out, with the scope undecided and no clarity on when, or whether, any rules would take effect. This is a report of internal deliberations, not a policy.
Why China might do it
The logic, as described, closely tracks Washington's. The United States has spent years restricting China's access to advanced chips and chipmaking tools on national-security grounds. Beijing appears to be weighing the mirror image: treating its own leading AI models as a strategic asset to be guarded rather than a product to be spread. Concerns reportedly include foreign militaries using the models or probing them for weaknesses, and a broader instinct, now that Chinese AI is near the cutting edge, to protect an advantage rather than broadcast it.
That would mark a sharp turn. Chinese open models, most prominently DeepSeek's R1 earlier on, won global users precisely because they were free, capable and easy to adopt, undercutting costly Western systems and unsettling Silicon Valley. Curbs would start to close that door.
Why it matters
For the global AI industry, the report points to a further hardening of the split between the American and Chinese technology worlds. If China restricts exports of its models much as the US restricts exports of its chips, companies and developers elsewhere may increasingly have to choose one ecosystem or the other, rather than mixing the best of both. Widely used Chinese models could become harder to obtain outside China, nudging international users toward Western alternatives and fragmenting a field that until now has been strikingly borderless.
For investors, it is a reminder that AI's biggest risks are not only technical but political. The value of an AI model depends partly on who is allowed to use it, and that, increasingly, is a decision governments intend to make. For now, though, this remains a report of what Beijing is considering, not what it has done, and the distinction matters.
