China Approves DeepSeek and MiniMax to Deploy Models Overseas Under Its New Export Rules

China has cleared DeepSeek and MiniMax to deploy their AI models overseas, Reuters reported on October 23, making them the first labs registered under the country’s new AI export-control regime, per a state-media-affiliated entity. The clearance operates inside the framework Beijing proposed earlier this month: export controls covering 23 AI technologies, including LLMs, chip designs, and agent systems, with restrictions on deploying models containing restricted technology abroad. In effect, China is building a licensing layer between its labs and the global developers who download their weights.

The mechanism, and what it licenses

Read the two halves of this story together. On October 16, Techmeme-carried reporting described China’s proposal to add 23 AI technologies to its export-control list, framing it as a way to prevent unauthorized transfer of sensitive capabilities. One week later, the first two approvals arrive, and both names are strategic: DeepSeek, the open-weights lab whose V4.1-Flash refresh we covered at release, and MiniMax, whose M3.1 Flash launch revealed 200 trillion tokens of anonymous production traffic. The Chinese state is not restricting its open-weights champions; it is credentialing them, converting what was an informal free-for-all into a state-supervised channel where global distribution is a granted privilege rather than a default.

What this changes for the open-weights ecosystem

For the millions of developers running Chinese open models, nothing changes today: the approved models are the ones already widely deployed. What changes is the legal architecture underneath. Until now, anyone could pull a DeepSeek checkpoint from Hugging Face with no more friction than any other download. Under the new regime, future releases from Chinese labs will pass through a state approval step, which creates three consequences. First, release timing becomes a policy instrument; a lab awaiting clearance cannot ship. Second, the open-weights supply chain gains a compliance question for Western enterprises: a model’s weights now carry a licensing provenance from Beijing, which procurement and legal teams will need to assess alongside the US government’s own exclusions. Third, the US and China have now symmetrically weaponized the same tool: Washington excludes models from procurement, Beijing licenses models for export. The open internet of weights is being governed from both ends.

What to watch

Three signals will show how hard the new regime bites. First, the approval pipeline’s speed and selectivity: if future DeepSeek, MiniMax, Qwen, GLM, and Kimi releases clear quickly, this is theater; if releases stall or ship in censored configurations abroad, it is control. Second, whether Hugging Face and other distribution platforms begin marking models with export-approval status, since the platforms are where the licensing question becomes operational. Third, the reciprocal question in Washington: the same week this clearance arrived, reporting described OpenAI nearing a $30 billion raise at a $1.25 trillion valuation (CNBC), and US policymakers will read a state-coordinated Chinese model export pipeline as exactly the national-champion behavior their own industrial policy mirrors. The open-weights era was defined by the absence of permission; both governments are now in the business of granting it.