POSTS FROM

News

AI ★ FEATURED

China's GLM-5.2 matches a banned US AI on cybersecurity tasks and there's no export order that can stop it

Zhipu AI's GLM-5.2, released under the MIT license on June 13, has matched Anthropic's restricted Mythos model on independent cybersecurity benchmarks from Semgrep and Graphistry. Because the model is freely downloadable with no regional restrictions, US export controls designed to contain Mythos's vulnerability-detection capabilities have no mechanism to reach it. Z.ai co-founder Tang Jie says a full Fable-class Chinese model is coming before Q1 2027.