China Gave Away an AI That Matches the Model the US Just Banned

On June 12, 2026, the US Commerce Department issued an export-control directive barring Anthropic from supplying its Fable 5 and Mythos 5 models to any foreign national, forcing the company to disable both models worldwide outside the US. The trigger was a narrow jailbreak technique that allowed users to bypass the models' built-in restrictions.

China's response — one day later

On June 13, Beijing-based Z.ai (formerly Zhipu AI) released GLM-5.2 — an open-weight model under a permissive MIT license, trained entirely on Chinese Huawei Ascend chips with no Nvidia hardware involved.

Independent tests found parity

Two independent security evaluations published in July found that GLM-5.2 has matched or approached leading US models on the exact class of cybersecurity capability that justified the export ban on Fable 5 and Mythos 5. Within a week, GLM-5.2 climbed to the top of the open-weight leaderboards, and Z.ai's market valuation passed HK$1 trillion (roughly $128 billion).

What this means in practice

The most capable model many users outside the US can now legally access turns out to be a free download from a company that formally sits on Washington's trade blacklist. Export controls work by restricting access to specific models — but when a competing open model reaches comparable capability independently, the ban stops containing the spread of the technology itself. What it still determines is simply which company captures the commercial upside.

This material is for informational purposes only.