China is considering restricting international access to its leading AI models.

      China's open AI models have been a boon for developers worldwide, but officials in Beijing may be considering reining them in. According to a report by Reuters, discussions have taken place regarding restrictions on who outside the country can access China's top AI models. Over the past month, the Ministry of Commerce has conducted these meetings involving companies like Alibaba, ByteDance, and the startup Z.ai. The conversations focus on the most advanced models, some of which have yet to be released.

      What's being proposed

      The plans extend beyond a straightforward export ban. They include regulations on open-weight models, the freely available systems that have contributed to the international popularity of Chinese AI, as well as closed models. Notable examples include Alibaba's Qwen, ByteDance's Doubao, and Z.ai's GLM-5.2.

      Two additional proposals have emerged. One suggests classifying the leak or theft of proprietary AI technology as a national security offense, while the other aims to restrict which investors can finance domestic AI companies. However, sources indicate that no decisions have been made yet, that any restrictions may only apply to future models, and there is currently no timeline for implementation.

      Details on the proposed restrictions remain unclear. A group of Chinese legal scholars has proposed a tiered approach: a simple filing process for basic tools, security evaluations for more advanced systems, and a domestic-only restriction for the most sensitive models.

      Significance of the development

      This potential shift represents a significant change. China gained international goodwill by making its models widely available, with European developers relying on affordable open weights from companies like DeepSeek as a substitute for expensive American systems. Introducing restrictions would reduce this supply, and Reuters highlights that costs could increase for numerous businesses dependent on these models.

      This development reflects a trend similar to actions taken by the U.S. The U.S. has implemented measures to prevent China from replicating its models and recently imposed restrictions on Anthropic’s advanced systems due to security concerns, mirroring what Beijing now apprehensively perceives in reverse. China has already established barriers, grounding its AI researchers and controlling who can support its startups. Treating AI models as state properties would represent the next step in this trend.

      It remains uncertain if any of these proposals will eventually become law. If they do, the era of freely downloadable Chinese AI could quietly come to an end.