Media365
A draft proposal by the Ministry of Commerce of the People’s Republic of China (MOFCOM) adds advanced artificial intelligence technologies to its export-control list. The restrictions seem to be designed more as a switch than a wall. Almost nothing changes for exporters today, but China now holds the means to monitor, slow or stop the flow of its AI technology abroad whenever it chooses.
On October 16, MOFCOM issued two things together. The first is a draft revision of the Catalogue of Technologies Prohibited or Restricted from Export, adding 23 AI items within a wider list of 40 technologies (including biotech and solar). Once in force, restricted items will need an export licence and prohibited items cannot be exported. The second is a temporary registration requirement, already in force, under which exporters must disclose any transfer of covered technology (model weights, code, architecture, hardware optimisations and so on) before it happens. Both apply to transfers of technology, not to the sale of AI services.
The order does four things at once. First, it maps outbound technology flows. Every technology transfer covered under its ambit must now disclose the technology, the parties, the destination, the end use and the value. Second, it creates a veto on technology transfer while promising not to use it. That route can be narrowed or withdrawn in the future. Third, the order claims to be motivated to address cyber safety concerns, but the coverage is much broader. Lastly, it mirrors US leverage as Beijing now has a matching toolkit over models, agents, and chip designs.
An important inclusion in the restricted list is foundation models and related core technologies that meet one of many thresholds such as 10^26 training FLOPs, 2 trillion-plus parameters, a top-10 global ranking on a qualifying measure, advanced agentic capabilities and training on core or “important” data as defined under China’s Data Security Law. Open-weight models that have been adopted across the world get “automatic licensing”, which means that a licence is still legally required, but approval is meant to be quick and procedural.
The design choices in this order make more sense as instruments of control and leverage, and not just safety measures. The following arguments back up this reading.
First, the triggers for restriction are many and form a wide and discretionary net. The restrictions seem to target capabilities rather than its riskiest systems, as indicated by criteria such as presence in global top 10 rankings or training on important data. This would catch almost any Chinese frontier model.
Second, the order does not restrict API access: Foreign users can use the same capabilities as a service. But that keeps them dependent on a Chinese provider, unlike a transfer of the model’s weights, which would hand over control. Beijing has left open the channel that preserves its leverage and licensed the one that gives leverage away.
Third, the change-of-control clause targets relocation. Registration covers deals where foreign ownership or control changes, even if no code crosses a border. This is likely aimed at Chinese AI start-ups moving their legal home abroad, as Manus did when it relocated to Singapore and was later acquired by Meta.
Finally, the other controls target economically valuable technologies, not just the high-risk ones. Chip design at 14nm and below, advanced packaging, 10,000-accelerator clusters and cross-vendor chip scheduling carry no misuse risk. Cross-vendor scheduling is China’s workaround to US chip controls, for example, mixing Nvidia and Ascend accelerators in one training run. Solar and biotech also in the same package indicate that this is a competitiveness instrument with a safety and risk framing.
The order changes the terms on which Chinese AI technology reaches India. Read alongside Washington’s actions this year, it also shows that both AI superpowers now want to exercise control over the technology the rest of the world builds on. Three implications follow from that.
First, nothing changes now, but the basis for leverage exists. Developers fine-tuning Qwen or DeepSeek are unaffected, as already released weights are in the public domain. Future releases that cross the thresholds, however, will need a licence that Beijing can control.
Second, diversity of jurisdictions matters as much as diversity of model vendors. China has now built a licensing gate for open releases, and the US has shown this year that it can cut off access to closed frontier models at a few days’ notice. A resilient open ecosystem needs serious contenders from other states; a consortium of companies and governments from India, Europe, and the Gulf could be a viable pathway.
Finally, avoid lock-in to a single model or agent framework. Indian builders should design technology stacks that prioritise portability and interoperability. India’s opportunity lies in diffusion or widespread adoption of the technology. This requires investing in the talent and compute to fine-tune, distil and adapt whichever open model is best and available.
The writer is associate fellow, Takshashila Institution
