AI Risk Clock
Doomsday Clock5 min to midnight
→
🌑 Dark edit
bloomberg.com2026-10-23

DeepSeek, MiniMax Win Approval for Global Use, State Media Says

PolicyModelsIndustry

Beijing has decided that DeepSeek and MiniMax Group may take their models abroad, and it has conveyed this by way of Yuyuantantian — a social account affiliated with China Central Television, the state broadcaster. Not a ministry order, not a stamped PDF, not a docket number: a post. One imagines the compliance department at MiniMax printing the screenshot, laminating it, and filing it under "approval, quite a lot of". The rest of us are left to deduce which models, which markets, and which regulator, because the post did not say. Bloomberg got the story from the post, which is how most of us now receive industrial policy: second-hand, cheerful, and without a footnote.

The post did, however, promise that much of Chinese AI will remain free for export, a sentence in which the word "free" is doing the sort of heavy lifting usually reserved for a forklift. Free as in no licence fee, free as in no friction, free as in you will discover what it costs you at some later date, when your own product is undercut by a model that arrived with a press release and no invoice. It also addressed Beijing's draft proposal for export control measures on AI tools and services — draft being the operative term, the regulatory equivalent of "we're thinking about maybe putting a lock on that door, once we've agreed what a door is".

Two labs, one line in a state-media post, and the open-weight map redraws itself overnight. The elegant part is the packaging: approval and export control announced in the same breath, so the measure that permits and the measure that restricts arrive as a matched pair. A cat flap installed alongside a promise to consider a burglar alarm. Everything shipping, nothing signed, and "much of Chinese AI" — that vague, generous, unenumerated "much" — heading for a server near you.

Read this story in another voice
● REC · 2026
DeepSeek, MiniMax Win Approval for Global Use, State Media Says — Dark edit · AI Risk Clock