DeepSeek and MiniMax have completed registration under China’s new pre-export filing rule for AI technology and have gone ahead with open-sourcing and overseas licensing, Chinese state media said on 22 October 2026. The account came from Yuyuan Tantian, which Pandaily describes as a social media account run by China Media Group, the state broadcaster; Jiemian News relayed it, and Bloomberg reported it on 23 October under the headline “DeepSeek, MiniMax Win Approval for Global Use, State Media Says.”
The rule is the temporary registration step that came with the draft revision of China’s Catalogue of Technologies Prohibited or Restricted from Export, which the Ministry of Commerce and the Ministry of Science and Technology opened for public comment on 16 October. The draft adds 23 AI technologies, covering large models, agents, computing clusters and information processing. The ministry’s spokesperson said that once the measures take effect, the great majority of AI technologies can still be exported freely.
Business Pill · OPEN WEIGHTS
A short Business Engineer explainer on open weights: renting a model versus running it yourself. It teaches the general idea only and says nothing about any company in this piece.
The key insight: As we read it, the new rule draws its line at where a model physically goes: according to the Ministry of Commerce, open-sourcing and online model services sit outside the controls or on a fast track, while local deployment of a restricted model abroad is controlled.
Update (23 Oct): Reuters picks up the registration, and a second post. DeepSeek and MiniMax “have won state approval for global use of their AI models,” Reuters reported on 23 October, citing the same state-broadcaster-affiliated account. Reuters added that in a separate post on Friday the account, Yuyuan Tantian, said overseas internet users had also urged other Chinese open-source large language models to follow suit, and that DeepSeek and MiniMax did not immediately reply to its requests for comment. For how the two listings traded this week, see our piece on MiniMax and Zhipu shedding $150 billion from their peaks.
Update, 29 Oct: which models the draft catches.Reuters reported on 28 Oct that the draft rules distinguish between foreign customers using Chinese AI services and foreign companies obtaining the models, data or expertise needed to operate or reproduce them independently. China is expected to adopt the proposals by the end of the year, according to a source familiar with the matter cited by Reuters.
The model tests sit in the draft list attached to the 16 October MOFCOM announcement. Its large-model entry (266504X), in the restricted rather than the prohibited part, covers a model’s weights, key checkpoints, source code and architecture, plus training-optimisation and inference-acceleration technology, if the model meets any one of three tests: training compute of at least 1026 floating-point operations, at least 2 trillion parameters, or a global top-10 place in public rankings by mainstream evaluators of overall performance, specialised capability, global call volume, safety or training efficiency. As Reuters notes, the draft does not say which rankings count.
Reuters also reported that researchers at major Chinese AI labs are seeking clarification on whether they can train or run their most advanced models on foreign-owned overseas servers, according to a person briefed on the matter. An analysis by law firm Fangda, cited by Reuters, says the registration requirement covers qualifying transactions already underway; the announcement itself says exports that have started but not finished fall within the registration scope.
The Yuyuan Tantian text, as relayed by Jiemian News on its Sina Finance account, reads: “ๆฎๆ๏ผๆฟ็ญๅ ฌๅธไปฅๆฅ๏ผDeepSeekใMiniMaxไธคๅฎถๅคด้จไผไธๅทฒ้กบๅฉๅฑฅ่ก็ป่ฎฐๆ็ปญ๏ผๅนถๅฎๆๅผๆบๅๆตทๅคๆๆไฝฟ็จใ” In our translation: it is understood that since the policy was published, DeepSeek and MiniMax, two leading companies, have smoothly completed the registration procedures and completed open-sourcing and overseas authorized use.
The same text argues that the catalogue’s core is “ๅๅฎ่พน็๏ผ่้ๅ็บฏ้ๅถAIๆๆฏๅฏนๅคๅบๅฃ”, which in our translation means drawing boundaries rather than simply restricting AI technology exports. It notes that the ministry’s answers say more, and more specifically, about what is not covered than about what is.
Bloomberg’s headline speaks of approval. The state-media text describes completed registration procedures; the registration step asks exporters to record basic information about themselves and the recipient, the destination country or region, the technology and its use, and the expected export amount, according to the announcement.
Of the 40 technologies China’s draft revision would add to its export-control catalogue, 23 are AI technologies, 12 photovoltaic and 5 biomedical, according to the Ministry of Commerce and Ministry of Science and Technology announcement of 16 October 2026.How the 40 proposed additions split. Source: MOFCOM and MOST Announcement 2026 No. 48 and the ministry spokesperson, via Xinhua, 16 October 2026.
What Was Posted on X
DeepSeek and MiniMax registered under Chinaโs new AI export rules, then kept open-sourcing models.
Since October 16, China requires AI companies to register with the government before sending covered AI technology abroad, and publishing a modelโs weights or code for anyone in the world to download (open-sourcing) counts as that kind of export, so even open-source releases now need a filing first.
Rohan Paul’s 23 October post: DeepSeek and MiniMax registered under China’s new AI export rules and then kept open-sourcing models; since 16 October, he writes, publishing a model’s weights or code for anyone to download counts as an export, so even open-source releases now need a filing first.
DeepSeek and MiniMax registered under Chinaโs new AI export rules, then kept open-sourcing models.
Since October 16, China requires AI companies to register with the government before sending covered AI technology abroad, and publishing a modelโs weights or code for anyone in the world to download (open-sourcing) counts as that kind of export, so even open-source releases now need a filing first.
Rohan Paul’s 23 October post: DeepSeek and MiniMax registered under China’s new AI export rules and then kept open-sourcing models; since 16 October, he writes, publishing a model’s weights or code for anyone to download counts as an export, so even open-source releases now need a filing first.
What Beijing Proposed
Scope. The draft adds 40 technologies in three fields: 23 in AI, 5 in biomedicine and 12 in photovoltaics, according to the announcement published by Xinhua.
Prohibited and restricted. Of the 40, 14 are proposed for an export ban and 26 for restriction, Xinhua reported. The ban list includes AI-driven malicious code generation and mutation technology; the restricted list includes AI-assisted vulnerability mining.
Registration now. Until the revised catalogue takes effect, exporters must register on a Ministry of Commerce platform before exporting any of the listed technologies, the announcement says.
Comment period. The public can comment from 16 October to 15 November 2026, according to the announcement.
The Data Point: What Counts as an Export
The ministry’s answers sort the ways Chinese AI reaches users abroad into what is controlled and what is not. Set out side by side, open-sourcing and online model services sit outside the controls or on a fast track, while local deployment of a restricted model abroad is controlled.
How Chinese AI goes abroad
Treatment under the draft, per MOFCOM
Technology already public through open source or academic papers
Not included in export management
Open-sourcing AI technology; Chinese firms’ own overseas compute centres; chip tape-out abroad; joint R&D using overseas resources; international standard-setting
Automatic licensing and other facilitation measures, which the ministry says basically achieve “ๅณๆฅๅณๅบ” (export upon filing, our translation)
A restricted model offered abroad through API calls, apps or web pages, SaaS, or lightweight on-device deployment
Not within the scope of large-model technology export management
A restricted model deployed locally overseas
Within technology export management
A restricted model deployed on the company’s own overseas cloud
Automatic licensing and other facilitation measures may apply
Source: the Ministry of Commerce spokesperson’s answers and the 16 October 2026 announcement, as published by Xinhua; English wording is our translation of the Chinese text.
What the Two Companies Have Done
Pandaily, citing Huxiu, reported that DeepSeek open-sourced a large-scale inference framework called DeepInfer and a companion toolkit, deepseek-recipe, on 20 October. DeepSeek’s GitHub organisation lists DeepInfer, described as a “High-performance, large-scale inference framework for DeepSeek models.”
Pandaily also reported that MiniMax has set up a formal application route for regions its earlier model license excluded, including the United States, the European Union, the United Kingdom and South Korea, under which organizations there need to pass a compliance review and obtain a license before deploying the models locally. Separately, MiniMax wrote on X on 23 October that “M3.1 Flash weights are now open,” with Together AI as a day-0 launch partner; the post does not mention the registration.
Connecting the Coverage
Our July export-curbs piece reported, on 7 July 2026, a Reuters exclusive that Chinese authorities had held meetings with Alibaba, ByteDance and Z.ai to discuss potentially restricting overseas access to China’s most advanced AI models, and that participants had proposed a tiered regime. Our open-weight enforceability piece followed on 21 July, describing the idea as consideration, not policy. What’s new: the draft now exists, with a registration step in force, and two named model makers have been reported to have gone through it.
Our DeepSeek and Huawei piece reported on 22 October 2026 that DeepSeek raised about 100 billion yuan ($14.98 billion) in an equity round, according to Bloomberg reports relayed by Reuters and The Edge Malaysia. Our Treasury piece reported that the US Treasury Department on 22 October 2026 proposed a rule that would bar US investors from backing companies in China and other countries of concern that develop any AI system. By our count, the Chinese announcement and the US proposal were published six days apart.
The Structural Read
As we read it, the state-media account gives the rule a first working example: two of the best-known Chinese model makers are reported to have completed the registration and gone ahead with open-sourcing and overseas licensing.
As we read it, the ministry’s line leaves open weights and online access outside large-model export management or on automatic licensing; Pandaily writes that open weights, open code and API access are expected to continue largely as before.
As we read it, the controlled part is narrower and more specific: the ban list includes AI-driven malicious code generation, and the restricted list includes AI-assisted vulnerability mining, according to Xinhua.
Bloomberg, 23 October 2026
“Much of Chinese AI will remain free for export”
Three Implications
FOR CHINESE MODEL MAKERS Open-sourcing qualifies for automatic licensing, which the ministry says basically achieves export upon filing; local deployment of a restricted model abroad falls within export management.
FOR FOREIGN USERS API calls, apps, web pages and SaaS built on a restricted model are not within large-model export management, according to the ministry; Pandaily reports, citing Huxiu, that MiniMax now asks organizations in some regions to pass a compliance review before local deployment.
FOR THE COMMENT PERIOD The public can comment until 15 November 2026, and Pandaily notes the final scope could change before it takes effect.
The framework’s core pattern: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize)”.
As we read it, Beijing’s draft applies a version of that pattern at the level of a country: open-sourced models and online services stay open, while a short list of capabilities, such as AI-driven malicious code generation, is proposed for a ban. The framework was published in March 2026 and does not discuss China’s export catalogue.
What Is Not Established
We have not seen DeepSeek’s or MiniMax’s registration records, and we did not contact either company. The account of what they did comes from state media and from Pandaily citing Huxiu. The catalogue is still a consultation draft; its final scope could change after the comment period. Our translations of the Chinese texts are ours. Nothing here is investment advice.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This piece is part of this week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Chinese state media say DeepSeek and MiniMax have completed registration under the pre-export filing rule that came with China’s draft list of 23 AI technologies, and have gone ahead with open-sourcing and overseas licensing. The Ministry of Commerce says open-sourced technology and online model services fall outside the controls, while local overseas deployment of a restricted model is within export management.
A note on sourcing. This piece is based on the Chinese state-media account published by Yuyuan Tantian and relayed by Jiemian News, the Ministry of Commerce and Ministry of Science and Technology announcement of 16 October 2026 and the ministry spokesperson’s answers as published by Xinhua, Bloomberg’s report, Pandaily’s report, DeepSeek’s GitHub listing, MiniMax’s post on X and our own earlier pieces. Translations from Chinese are ours. Nothing here is investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
