No. 94
2026-10-21
the AI pulse — brewed, not raw
 AI CRITIQUE
Görsel, bina ile kilit, kalemli bir belge ve bilgisayar devre kartını birleştiren stilize bir illüstrasyonu göstermektedir; bu da veri güvenliği ve teknolojiyi temsil etmektedir.

China adds 23 AI technologies to export control list

Read the article    English (automatic translation)

Summary

China has launched a public consultation process, open until November 15, regarding export restrictions covering 23 artificial intelligence technologies. The scope includes large models, agents, computing clusters, and data processing. Open-source weights and API-based access are explicitly excluded from the regulatory regime.

Why it matters

This move represents a concrete step toward a Chinese AI export control regime that logically mirrors the US chip restrictions. It directly impacts how companies in Europe and other countries deploy China-originated frontier models on-premise. The distinction between API access and local deployment creates a two-tiered market that will shape corporate AI procurement and cloud strategies on both sides of the Atlantic. Additionally, Washington's October 20 order, which excludes DeepSeek from federal civilian agencies, demonstrates a coordinated tightening of AI access on both sides.

Highlights

  • 23 AI technologies are targeted; 14 are subject to a complete ban, 26 require a license; prior notification has been mandatory since October 16
  • According to MOFCOM, open-source weights and API/SaaS calls for DeepSeek, Qwen, and GLM are explicitly out of scope
  • Reported thresholds: ≥10²⁶ FLOPs for training, ≥2.000 billion parameters, or ranking in the top 10 in global benchmarks
  • Deploying a restricted model on-premise abroad is subject to export controls; 'foreign own cloud' may qualify for an automatic license
  • Washington's October 20 order excludes DeepSeek from US federal civilian agencies

Critical take

The article relies on a single third-party summary by Geopolitechs for the technical details of the 23-item annex, as direct access to the original list was not possible. The 'global top 10 ranking' criterion is so ambiguous that it can be manipulated at will by the party defining the benchmarks. The text bases the critical boundary between API and deployment on a single word in a Q&A session. This makes it difficult for readers without access to the underlying legal text to verify the scope.

Bu kritik, yerel yapay zeka modeli Qwen3.8-27B tarafından yazılmıştır. Haberi ben seçtim; kritik metnini yayımlanmadan önce tek tek denetlemiyorum.This critique was written by the local AI model Qwen3.8-27B. I chose the story; I do not review each critique before it is published.
Back to the issue