For much of the past two years, China’s approach to artificial intelligence looked like the opposite of a walled garden. Its labs shipped models around the world at breakneck speed, and cheap, freely downloadable Chinese systems became a lifeline for small teams and businesses that could never afford the pricey frontier models coming out of the United States. Now Beijing appears to be reconsidering that openness.
Chinese authorities have spent the past month in talks with the country’s biggest tech firms about restricting overseas access to its most advanced AI models, including systems not yet released, according to three people familiar with the discussions who spoke to Reuters. The Ministry of Commerce convened the sessions, and the guest list was telling: Alibaba, ByteDance, and the startup Z.ai were all in the room.
The talks signal a shift in how China thinks about the technology. For years the priority was reach and market share. Now, like Washington, Beijing is starting to treat cutting-edge AI as a strategic national asset that needs guarding rather than a product to give away.
Several ideas were reportedly on the table. One would classify the unauthorized disclosure or theft of proprietary AI as a violation of national security law. Another would tighten control over which investors are allowed to back homegrown AI companies, building on earlier moves that already required firms such as Moonshot AI and StepFun to get government approval before accepting U.S. funding. A panel of Chinese legal scholars has floated a tiered system: a light registration for basic tools, security reviews for more capable ones, and a domestic-only lockdown for the most sensitive models.
Nothing has been decided. The sources cautioned that any restrictions might apply only to future models, and that no timeline exists. But even the discussion marks a notable turn for a country that earned considerable global goodwill by making its models widely available.
The irony is that Beijing’s move mirrors Washington’s. The United States has taken steps to stop China from copying its systems and recently restricted access to Anthropic’s most advanced models on security grounds, precisely the kind of measure China now appears to fear in reverse. Part of Beijing’s concern, according to the reporting, is that advanced Western cybersecurity tools could be turned against Chinese systems, prompting calls at home to build a comparable capability.
For the wider world, the stakes are practical rather than abstract. European developers in particular have leaned on inexpensive open-weight Chinese models as an alternative to expensive American ones. If Beijing narrows that supply, costs would climb for the many businesses that have quietly built their operations around affordable Chinese AI. Scottish and European tech firms working on tight budgets could feel the pinch as much as anyone.
China has already built parts of this wall. It has kept its AI researchers close to home and taken a firmer hand in deciding who can fund its startups. Treating the models themselves as protected state assets would be the next brick. Whether any of it becomes law is still unclear, but if it does, the era of freely downloadable Chinese AI may quietly come to an end.
