China weighs restricting overseas access to advanced domestic AI models

If Beijing does move to restrict overseas access to its strongest models, it would mirror the U.S. decision to cut off Anthropic’s Fable and Mythos lines for foreign users—turning model access itself into an instrument of economic statecraft. For now this is at the discussion stage, but the fact that Alibaba, ByteDance, and frontier players like Zhipu AI are in the room shows the idea is being taken seriously. The immediate impact would be on developers and companies outside China that have come to rely on cheap, capable Chinese models as an alternative to U.S. APIs.

More broadly, this would accelerate the balkanization of the AI stack: chips have already been split by export controls; models would be next. That could slow global diffusion of near‑frontier capabilities, but it won’t stop capability growth inside the major blocs. Instead, it incentivizes each side to over‑invest in self‑reliance across chips, models, and cloud. For the race to AGI, that likely means more money and talent flooding into parallel, semi‑walled ecosystems, with less cross‑pollination but also less shared visibility into emerging risks.