China Wants Its AI to Go Global, on Its Own Terms

The new regulations introduce a pre-export registration requirement for a range of AI technologies. These include core components of advanced models, such as weights and source code, as well as AI agents, AI chip design, and the interconnect and scheduling technologies for clusters exceeding 10,000 accelerators. This registration is triggered by a transfer of ownership abroad, a license for overseas use, or a change of owner or actual controller to a foreign entity.

This regulatory shift aims to provide Chinese authorities with advance notice of transactions involving strategic AI technologies. Unlike the previous "Manus controversy," where Beijing intervened to block Meta's acquisition of an AI agent startup after the deal had closed, the new rules require registration before a transaction is finalized. This approach also signals that relocating a company's headquarters abroad will not automatically exempt listed technology from Beijing's oversight.

While the formal scope of prohibited or restricted technologies is described as narrow, with most AI technology remaining freely exportable, the controls target foundational elements. MOFCOM has clarified that offering restricted large models via APIs, apps, web interfaces, software-as-a-service applications, or lightweight on-device deployments falls outside the scope of large-model technology export controls. Additionally, open-source releases or academic publications of restricted technology are eligible for automatic licensing, as is deploying a restricted model in a Chinese firm's own overseas data centers.