Summary
On 16 October 2026, China's Ministry of Commerce and Ministry of Science and Technology published a draft, Announcement No. 48, proposing to add 40 technologies to the catalogue of technologies banned or restricted from export, 23 of them related to artificial intelligence.
The AI list covers advanced models, AI agents, computing clusters and chip design technologies. Models with at least 2 trillion parameters fall within the proposed controls. Technologies in the public domain, including open-source systems, are exempt.
An interim registration requirement took effect immediately and public consultation runs until 15 November 2026. The draft distinguishes between exporting the technology itself and providing services built on it, leaving room for Chinese firms to keep serving overseas users through APIs and cloud platforms.
WHY IT MATTERS FOR UPSC & STATE PCS
China has proposed export controls on 23 AI technologies, including large models and their weights, in a shift from controlling hardware to controlling AI capability. The move matters for countries, including India, whose developers increasingly build on Chinese AI models and cloud services.
Standard News
The Recipe Stays Home: How China Is Turning AI Into a Controlled Export
Here's what's actually happening. A modern AI model is two things. One is a service you can use: you type a question into an app or send a request through an API and an answer comes back. The other is the model itself, above all its weights, the billions of numbers learned during training that make the model work.
Whoever holds the weights can run, copy or modify the model without asking anyone. China's draft export controls, published on 16 October 2026 as Announcement No. 48, draw the line exactly between those two things.
The mechanism:
sell the meal, keep the recipe Think of a restaurant. Customers anywhere can order the dish. What the restaurant will not hand over is the recipe, the kitchen and the trained chefs. The draft does something similar. It proposes adding 23 AI technologies to China's export control catalogue, including:
- advanced models, with a threshold of 2 trillion parameters - AI agents that act with limited human direction - computing clusters - chip design technologies Exporting the technology itself would require registration or approval. Providing services built on it, such as API access or cloud applications, largely stays outside these controls. Technologies already in the public domain, including open-source models, are exempt.
Where the analogy stops working: a recipe, once published, can still be copied. Open-weight models already released remain usable worldwide. The controls bite on the next frontier models and on the know-how to build and run them.
Why Beijing is
doing this now China is adopting the logic of the US playbook it has been subject to. Washington spent years restricting advanced chips and chipmaking tools. A 2023 US executive order, since rescinded, also set a computing threshold for reporting on frontier models.
Beijing used a similar approach in December 2023, when it banned exports of technology for processing rare earths: sell the product, keep the capability. The draft applies that template to AI. Chinese firms can still earn global revenue from users, while the strategic core stays onshore.
Within days, state media reported that DeepSeek and MiniMax had secured open-source licences, showing how companies are already navigating the new rules.
Where
India stands Many Indian developers build applications on top of foreign models, including Chinese open-weight ones. Today's open models remain available. But the most capable future Chinese models may arrive only as services, priced and governed from Beijing and subject to change.
That is the strategic case for the IndiaAI Mission, approved in March 2024 with ₹10,371.92 crore to build shared computing capacity of over 10,000 GPUs. Access through an API is convenience. Owning compute and weights is sovereignty.
For GS3, this is the new shape of technology geopolitics: the contest has moved from who can buy chips to who is allowed to hold the model itself.
Quick Facts
Key numbers & takeaways — revise these first
-
China's Ministry of Commerce and Ministry of Science and Technology released the draft as Announcement No. 48 on 16 October 2026.
-
The draft proposes adding 40 technologies to China's export control catalogue, 23 of them AI-related.
-
Covered AI technologies include advanced models, AI agents, computing clusters and chip design.
-
AI models with at least 2 trillion parameters would be covered.
-
Technologies already in the public domain, including open-source systems, are exempt.
-
Public consultation on the draft runs until 15 November 2026.
-
The US Bureau of Industry and Security tightened controls on advanced AI chip exports to China in October 2023.
-
The EU AI Act treats general-purpose models trained with more than 10^25 FLOPs as posing systemic risk.
-
The IndiaAI Mission was approved in March 2024 with an outlay of ₹10,371.92 crore to build sovereign compute capacity.
Connect the dots for your UPSC preparation.
Standard news covers the event. Log in to read our comprehensive analysis and uncover the hidden constitutional, structural, and ethical dimensions of this topic:
A plain-language breakdown of the four categories China wants to control and why model weights matter more than chips.
How China's draft mirrors the US approach of thresholds and capability controls and where the two diverge.
The December 2023 rare earth processing ban as a case study in Beijing's "sell the product, keep the know-how" strategy.
What the controls mean for Indian startups building on Chinese open-weight models and the role of sovereign compute.
Included in this analysis
Join thousands of aspirants analyzing the news deeply.
Unlock Premium — Rs.699 AnnuallyDon't have an account? Sign up for free