Z.ai's Tang Jie Argues Against China's AI Export Controls

Tang Jie, the founder of Z.ai, just published a memo arguing that AI capabilities should stay open and widely accessible, a direct response to reports that Beijing is considering its own export restrictions on frontier models, the kind of controls the US imposed on Fable 5.

This matters because it's the first serious pushback from inside the Chinese lab ecosystem against what amounts to a policy mirror. The US government banned Fable 5 for 19 days in June. Now China's apparently weighing whether to lock down its own models from overseas users. Tang is saying: don't do that.

The structure of his argument is worth watching. He's not defending the US position. He's defending the principle that open access to AI tools drives faster iteration, broader innovation, and ultimately benefits the developers who build on top of these models. His company's GLM models have gained traction specifically because they were openly available. Restricting them cuts off that feedback loop.

What's happening here is the negative feedback from US policy. The Fable 5 ban was justified on national security grounds, a model that could write exploits was too dangerous to leave in the hands of foreign nationals. The justification was specific and narrow. But the precedent it set is global. If the US can restrict a frontier model on security grounds, why can't China? Why can't the EU?

The answer, from Tang's perspective, is that openness compounds. Every restriction creates pressure for symmetry. Every symmetry creates more restrictions. You end up with balkanized model access: US companies serving US users, Chinese companies serving Chinese users, and everyone losing the spillover effects that come from operating on the same frontier.

This is a hard argument to win politically. Governments like control. Labs like moats. And security officials like certainty. The appeal of an export control regime is that it's clear, enforceable, and makes a country feel like it's protecting itself.

But there's a second read here. Tang's memo might not be aimed at convincing Beijing. It might be aimed at other lab founders. It's a shot across the bow saying: if you go along with this, you're fragmenting the entire global AI market. The winners of that fragmentation won't be the countries doing the restricting. They'll be the AI companies that can operate in both ecosystems, and there are very few of those.

The US already faces the reverse problem. OpenAI, Anthropic, and Google are all working to figure out how to comply with export controls while staying global. The answer they're finding is: you mostly can't. So you shrink your addressable market, you get cut off from non-US data and feedback, and you move slower.

Tang's argument is that China doesn't have to repeat that mistake. And if Beijing is smart, it won't. But if it does, the fragmenting won't hurt Z.ai more than it hurts the US labs. It'll hurt both. It'll hurt the developers relying on both ecosystems. And it'll slow everyone down.

That's the real outcome Tang is fighting. Not China staying open, but the entire frontier slowing because two governments are playing symmetry games with the infrastructure everyone actually needs.