The Chinese government is looking to restrict or outright ban the export of 40 sensitive technologies, with a specific focus on preventing Western access to advanced AI breakthroughs.

In draft guidance published by the China’s Ministry of Commerce (MOFCOM) and Ministry of Science and Technology (MOST), Beijing laid out new requirements for Chinese AI firms to formally register for any request

The guidance draws a line between foreign users accessing Chinese AI services, including using large language models (LLMs) such as Moonshot AI’s Kimi or DeepSeek’s V4 Flash, and directly exporting the underlying technologies for such systems.

The move places 23 AI technologies to China’s Catalogue of Technologies Prohibited or Restricted from Export, proposing an outright ban on the export of LLMs trained on sensitive data set out in China’s Data Security Law as well as any agentic systems intended for national security or offensive cybersecurity purposes.

Other AI exports will be assessed according to new criteria, including whether a model has two trillion parameters or if it is in the top ten worldwide in certain benchmarks. Open-source AI models released prior to the new rules are understood to be exempt.

Designers of AI chips will also be required to seek government approval for export licences of chips that are 14 nanometres or below, have high-speed interconnect capabilities, or are used for autonomous processes such as driving or computer vision.

Citing a source with information close to the matter, Reuters reported that Chinese AI developers are now seeking clarification over whether the new rules would prevent them from running their frontier models on overseas infrastructure.

Under the current regime, developers such as DeepSeek and Moonshot use train and inference models on cloud infrastructure hosted in regions such as Japan, Singapore and Thailand

The proposals, which also cover biopharma and photovoltaic energy technologies, are open to public consultation until 15 November.

This is the latest move by the Chinese government to safeguard its domestic technology firms.

Beijing blocked Meta’s $2 billion acquisition of the Chinese agentic AI developer Manus in April, with the Financial Times reporting that the firm’s founders had been blocked from leaving the country.

In August, it revised regulations that cover the intellectual property rights of Chinese AI chip designers, including expanded protections for those who design chips that have “photonic, quantum, or other functions”.

---