The Chinese government is drafting new regulations to control the export of advanced artificial intelligence models. The initiative targets models ranked among the top ten globally, as well as those with at least 2 trillion parameters. Authorities aim to restrict the transfer of model weights, source code, and architecture to ensure national security.
The proposed export control list includes 23 categories of AI-related technologies, such as AI agents, semiconductor design documentation, and computing clusters. Models trained on sensitive domestic data related to national security, the economy, or public interests will be subject to strict export prohibitions.
Companies are now required to register all international transactions involving these technologies, including those currently in progress. Deploying qualifying AI models abroad must be declared, while the operation of foreign computing infrastructure by Chinese providers will be eligible for automatic export licensing.
Officials from the Ministry of Commerce stated that these measures are intended to prevent the misuse of AI technologies in attacks against critical infrastructure or defense systems. Technologies already available as open source are exempt from these new regulatory requirements.
