China Drafts Export Controls on 23 AI Technologies Including Model Weights, Agents, and Chip Designs
China’s Ministry of Commerce and Ministry of Science and Technology jointly published Announcement No. 48 on October 16, proposing to add 40 technologies across three sectors to the country’s list of prohibited or restricted exports. Twenty-three of those technologies fall in AI.
The draft covers a wider scope than prior Chinese tech export controls, which focused largely on semiconductor manufacturing processes. This proposal targets the AI stack from model layer to silicon: large language models, AI agents, computing clusters, and chip designs are all listed. Restricted technologies include model weights and source code for products that exceed specified performance thresholds.
A public comment period runs until November 15. Importantly, the controls do not require that comment period to close before taking effect on a provisional basis: exporters of listed technologies must register with MOFCOM’s unified information platform immediately upon the announcement’s publication. That means the temporary registration requirement is active now.
What Is and Is Not Covered
The draft draws an explicit carve-out: technologies already in the public domain through open source or academic publication are not covered. That is a significant scope limit, particularly for AI model weights — a large share of frontier-adjacent Chinese models (Qwen, DeepSeek) have been released open-weight in recent years.
The 23 AI items break across four sub-categories: large language models, intelligent agents, computing clusters, and information processing. The 40-item broader package also includes 5 biomedicine technologies and 12 photovoltaics technologies.
Strategic Context
The proposal follows a string of AI-related security incidents — including unauthorized AI agent activity at several government sites globally — and comes as Beijing escalates its assertion of sovereignty over domestically developed AI infrastructure.
It also reverses what has been a consistent pattern from Chinese AI labs: publishing open-weight models internationally to gain adoption, mindshare, and safety research data from global usage. If the controls are implemented as proposed, exporting model weights for any model above the performance threshold would require government registration and, potentially, approval.
Zhang Yi, founder and chief analyst at iiMedia consultancy, told South China Morning Post the controls would introduce short-term adaptation costs for Chinese companies targeting international markets. Analysts noted a key unresolved question: exactly which open-weight models fall above the performance threshold — and therefore become subject to registration — has not been defined in the draft.
Implications for Global AI Competition
The draft controls are asymmetric in effect. US export controls have focused on chips and compute access, restricting Chinese labs’ ability to train at scale. China’s proposed controls would restrict the outbound flow of what those labs have trained — the model weights themselves. Used together by both governments, these two instruments could segment the global AI technology stack along national lines more sharply than any prior policy action.
The comment period closes November 15. If implemented without material changes, the controls would represent the most direct regulation of AI model exports by any major AI-producing country.