Proposed by China's Ministry of Commerce and Ministry of Science and Technology, Announcement No. 48 targets 40 technologies, including 23 artificial intelligence domains, under new export control rules. The restrictions capture model weights, code, and optimization know-how meeting thresholds like 10^26 FLOPs or 2 trillion parameters, while open-source releases remain exempt. Autonomous agents tailored for defense, critical infrastructure, and core government operations face absolute export bans. Although standard API access is permitted, deploying models locally abroad or licensing intellectual property to foreign entities requires regulatory oversight.
Key points
Providing remote services through standard application interfaces or web platforms does n…: Providing remote services through standard application interfaces or web platforms does not trigger model export restrictions.
Regulatory oversight applies to cross-border intellectual property licensing, transfers o…: Regulatory oversight applies to cross-border intellectual property licensing, transfers of control, and local overseas deployments.
Technologies placed into the public domain through academic publication or open-source di…: Technologies placed into the public domain through academic publication or open-source distribution are fully excluded from the rules.
Sources checked by our analysis pipeline: 1
Published: 2026-10-23
Categories: AI Governance and Regulation
China’s First Frontier AI Export Controls: MOFCOM Announcement No. 48
Key Takeaways
On October 16, 2026, China’s Ministry of Commerce (MOFCOM) and Ministry of Science and Technology (MOST) issued Announcement No. 48 of 2026. The announcement proposes adding 40 technologies to the Catalogue of Technologies Prohibited or Restricted from Export, 23 of them in artificial intelligence, and it imposes an interim pre-export registration requirement that is already in effect [1][2]. China’s catalogue has touched AI before: the 2020 revision added personalized recommendation and AI interaction interface technology [6]. Unlike those 2020 additions, the draft reaches frontier model weights, agent architectures, and AI-enabled offensive cyber capability, with explicit technical thresholds for each [1].
The announcement contains two instruments with different legal status. The catalogue additions are a draft open for comment until November 15, 2026, while the registration requirement applies now, including to exports that have started but are not yet complete [1][2]. Within the draft, model weights are restricted at frontier scale. Weights, key checkpoints, source code, architecture, and training and inference optimization…
The joint initiative brings together a domestic power producer, an investment firm, and a cloud provider to expand regional technological infrastructure. This collaborative effort addresses the rising computational demands of modern machine learning workloads by securing dedicated large-scale energy capacity.
Approximately 2,000 developers at the digital bank have switched to internal infrastructure to power their daily programming workflows. The transition slashed operational expenses to roughly one-tenth of previous external vendor fees, driving a 70% decrease in development lead time. By eliminating strict usage rationing, the company now treats open-weight systems as the standard baseline for all routine software tasks.
Legal scrutiny is mounting over tech companies utilizing media publications to train and power automated generative tools. An adviser to the European Union's highest court has warned that such practices could run afoul of regional intellectual property protections. This development signals significant regulatory hurdles ahead for technology firms deploying news material within their artificial intelligence ecosystems.