Morning Edition · Sunday, October 18, 2026Published at 2:13 AM EDT · New York
The draft exempts weights already made public and API access from abroad, but it does not address whether a Chinese lab's future open release above the threshold would itself count as an export. Public comments close November 15.

China's Ministry of Commerce and Ministry of Science and Technology on October 16 proposed adding 23 AI technologies to the national export control catalogue, including large models, agents, chip designs and computing clusters. The AI items sit inside a broader revision covering 40 technologies across AI, biomedicine and solar power, which Beijing justified on national security grounds. Public comments close on November 15.
The draft would cover models with at least 2 trillion parameters, or models that rank in the world's top 10 on tests of performance and safety, and it reaches weights, source code and other core components. According to a summary of the ministry text, technologies already made public through open-source releases or academic papers fall outside the rules. Access through APIs, websites, apps and software-as-a-service platforms is also exempt. A temporary rule that took effect on October 16 already requires a Commerce Ministry filing before a Chinese model's weights are sent abroad.
Most coverage reports the open-source carve-out as a blanket exemption. The text supports a narrower reading. The Decoder headlined the draft as tightening controls on open-weight models, and the Pillitteri summary notes that the exemption protects weights already released but says nothing about weights Chinese labs publish after the rules take effect. By that summary's account, Moonshot's Kimi K3 (2.8 trillion parameters) and Alibaba's Qwen3.8 (2.4 trillion) are exempt because their weights are already public. The text does not settle whether a successor of the same size could be posted to Hugging Face without a licence.
Total parameter count is a weak proxy for capability in mixture-of-experts models, so the top-10 ranking clause may do more of the work. DeepSeek's V4.1-Flash-1017, released with open weights on October 17, has a 552-billion-parameter backbone and falls well below the size line. None of the reporting reviewed names the leaderboard Beijing would use for the ranking test, which leaves the ministry discretion over which smaller models it captures.
The mechanism controls the artifact and leaves the service alone. A Chinese lab can keep selling API access to foreign customers. It would need approval to hand closed weights to a foreign cloud partner, to sell itself to a foreign acquirer, or to move its model into an offshore entity. China used the same tool in August 2020, when it added AI recommendation algorithms to the catalogue and gained an approval right over any transfer of TikTok's algorithm. The forced-sale talks over TikTok then continued for years.
Part of a tracked trend
AI Sovereignty and Export Controls on Frontier Models
Over the next 3-6 months, governments increasingly treat frontier AI models as strategic national assets — extending export controls to model access itself and backing domestic 'champion' labs as sovereignty plays.
Start a discussion in Townsquare.
More from this edition
Beijing would gain an approval right over transfers of model weights and over foreign acquisitions of Chinese AI labs. Chinese labs would keep their foreign application programming interface (API) revenue, while foreign buyers and investors in private Chinese AI companies would bear the licensing cost.
The core facts check out in the Ministry of Commerce notice. However, the ministry's own explanation says open-source publication of listed AI technologies will get automatic licensing, approved as soon as the filing is made. Future open releases therefore appear to need a filing rather than sit in an unaddressed gap. The ministry also says that deploying restricted models locally abroad counts as an export, and the Washington Post links the draft to United States policy while Beijing cites only national security.
An open-source-intelligence read of how likely this story is true with its real nuance, not a judgment of any outlet. It assesses the claim, weighing independent and adversarial reporting. How we label confidence.
What this means
The draft gives Beijing an approval right over the same thing Washington has sought to control: the transfer of frontier model weights and the compute designs behind them. That is direct evidence for the desk's thesis that AI capability is becoming a strategic asset nations compete to control, and it supports the two-superpower thesis, because only the United States and China are building parallel control regimes over model artifacts. The parties most exposed are foreign acquirers and investors in Chinese AI startups, Chinese labs planning offshore restructurings, and foreign clouds that want to host Chinese closed models locally. API revenue earned abroad is protected, so the direct hit to listed Chinese AI platforms is smaller than the headlines suggest. The likely loser is the price of private Chinese AI assets sold to foreign buyers, because a licence requirement on weights lowers what such a buyer can pay. The consensus reads the open-source carve-out as reassurance for the downloadable Chinese stack that many non-US users have adopted. The desk reads the text as giving Beijing a future veto over large open releases while it keeps the existing open-weight catalogue in circulation. The likelier outcome is that the final rule keeps the API exemption and the protection for already-public weights, and either leaves future open releases ambiguous or makes them subject to filing, within two to four months of the November 15 deadline. That outcome depends on Beijing still wanting Chinese open models to spread abroad as its main channel of distribution. The read is wrong if the final text explicitly exempts all future open-weight publication, at which point open release becomes the standard route for Chinese labs to avoid the regime.
What to watch
Observations to monitor, not financial advice.
Synthesized from: The Next Web · The Washington Post · Taipei Times · The Decoder · Pasquale Pillitteri · China AI Dispatch · Polylog editors
Comments
1Oct 18, 6:13 AM
Because the top 10 test is relative, a Chinese model could fall into or out of the catalogue purely as foreign models move in the rankings, with no change to the model itself.