DeepSeek and MiniMax Secure Export Registration Days After China Tightens Open-Source AI Rules

What happened

China's Ministry of Commerce (MOFCOM) and Ministry of Science and Technology published, in the previous week, a proposal to expand export controls to 23 technologies tied to artificial intelligence, including language models, AI agents and computing clusters. The measure also introduced an immediate official registration requirement for the covered technologies, according to the SCMP.

Days later, DeepSeek and MiniMax had already completed their export registration procedures, according to a report by the Yuyuan Tantian account, affiliated with state broadcaster CCTV, published on Thursday night. Both companies kept releasing open-source updates during the same period, which Chinese state media treated as a sign that the regulation isn't meant to close the door on exporting AI technology.

What the rule covers, and what it leaves out

The registration requirement applies to technology ownership transfers, overseas licensing agreements and changes in equity control involving foreign entities. Before closing this type of transaction, a Chinese company must register it with MOFCOM.

What's not included in the new requirement, according to a MOFCOM spokesperson quoted by the SCMP last week, is precisely the channel through which most developers access these models:

For the release of open-source technologies specifically, the Chinese government promised a faster process, called "automatic licensing," described by the MOFCOM spokesperson as essentially "immediate communication and export."

Why DeepSeek and MiniMax were already in a hurry

Both companies weren't sitting idle waiting for the regulation to be finalized. DeepSeek, based in Hangzhou, updated its V4.1 Flash model and open-sourced the code for its internal inference framework, a tool aimed at reducing the cost of running AI models. MiniMax, based in Shanghai, open-sourced the M3.1 model the Friday before the registration announcement.

According to Yuyuan Tantian, Chinese authorities began preparing the regulatory proposal in the first half of the year, consulting research institutes, industry associations and major companies before publishing the text. The stated intention is to separate two things: the transfer of AI technology itself abroad, which Beijing wants to restrict for national security reasons, and the provision of commercial AI services outside China, which the government says it wants to keep encouraging.

The backdrop: tech dispute with the US

The publication of the Chinese rule doesn't happen in a vacuum. It comes on the heels of US moves to further tighten Chinese access to American technology, including bills in Congress that would target the export of advanced chips and semiconductor manufacturing equipment.

During the same period, the US Department of Homeland Security ordered federal agencies to remove DeepSeek and third-party tools built on the model, citing national security risk. Beijing called the measure discriminatory.

Yuyuan Tantian made a point of marking the difference between the two approaches:

Simply put, AI companies can continue with their commercialisation efforts as usual. Once the relevant measures are officially implemented, the vast majority of artificial intelligence technologies will still be freely exportable.

Yuyuan Tantian, account affiliated with state broadcaster CCTV

The text also stated that China "is not following the path" of the United States of using export controls to isolate competitors and maintain global technological leadership, describing the review of its own rules as "a routine process carried out in accordance with laws and regulations."

What changes for those running these models in the US

For those who download open model weights like DeepSeek or MiniMax via Hugging Face and spin up their own server with vLLM, Ollama or similar, the direct news is reassuring: what's under registration is the transaction between companies (technology transfer, licensing, change of corporate control), not the public download of an open-source checkpoint. Access via API, official website and app is also explicitly outside the scope, according to MOFCOM.

But there's a point worth watching: the list of 23 covered technologies explicitly includes language models and AI agents, the two categories most used by engineering teams that adopted DeepSeek or MiniMax as a cost-effective alternative to closed models. The fact that the final text of the regulation hasn't been officially implemented yet, and that MOFCOM itself refers to "automatic licensing" as something to be applied in the future, leaves open whether this process will actually remain fast once the rule takes effect.

In short: nothing changes today for those already using these models outside China, but the regulation sets a precedent that future releases of open-source Chinese models may go through an extra layer of government registration before reaching the public, which is an additional geopolitical risk factor when choosing an AI stack based on these projects.

What remains open

The text of the regulatory proposal is still under consultation and hasn't been officially implemented, which means the registration process reported by state media may not exactly reflect the final procedure. There's also no public detail so far on exactly which licenses DeepSeek and MiniMax obtained, nor a set timeline for the definitive entry into force of the 23 proposed restrictions.