China has unveiled draft regulations aimed at restricting the export of certain advanced artificial intelligence (AI) technologies, reinforcing a strategy to retain core intellectual property and research within its borders while permitting limited foreign access to AI services. The proposed rules, released on October 16, are intended to bolster China’s competitive position amid increasing technological rivalry with the United States.
The draft distinguishes between foreign entities accessing Chinese AI through online platforms and those seeking to acquire the underlying AI models, data, or expertise necessary to operate or replicate them independently. While overseas users would generally continue to access Chinese AI tools via websites and cloud services, the transfer of core technologies or deployments abroad would require government approval.
This approach aligns with a business model favored by Chinese leadership and many technology firms, which emphasizes safeguarding intellectual property and research and development domestically while monetizing access internationally. Alfredo Montufar-Helu, managing director at consultancy Ankura in Beijing, noted that this framework "is the one the market was already converging on."
The draft regulations propose adding 23 AI-related technologies to a list that either bans or restricts their export. These include advanced AI models, autonomous agents capable of limited independent operation, and technology integral to chip design and operation of large-scale computing facilities. A related temporary requirement mandating registration of certain technology transfers took effect immediately.
The new rules are expected to influence how Chinese AI companies manage international collaborations, operate advanced AI models on foreign data centers, and handle foreign investments or ownership in AI assets. Several prominent Chinese AI firms, such as DeepSeek, Moonshot, and Z.AI, currently offer AI models to overseas customers through digital platforms. Some developers have relied on overseas computing facilities to access powerful chips, including those from Nvidia, which are restricted from export to China under U.S. sanctions.
Clarification remains sought by researchers at major Chinese AI laboratories regarding whether they can train or run their most advanced models on foreign-based servers. Officials and industry insiders warn that companies might be compelled to maintain separate domestic and international versions of certain AI models to comply with the regulations.
Legal experts involved in consultations with policymakers, including Liu Jing of Beijing Dacheng Law Offices, emphasize the need for clearer guidelines on the criteria used to determine a model’s inclusion under the export controls. The draft outlines three tests: computational workload exceeding 10^26 operations to train a model, model size surpassing two trillion parameters, or ranking within the world’s top 10 in specific performance or efficiency evaluations. However, the draft does not specify which ranking systems would apply.
China’s Ministry of Commerce has not publicly commented on the proposed regulations. The draft remains open for public consultation until November 15, with full adoption and implementation guidelines expected by the end of the year, according to sources familiar with the process. The measures come amid broader challenges faced by China’s technology sector, including stock market fluctuations and ongoing U.S. export restrictions on critical semiconductor technologies.