China AI developers publish safety tests for just 3.6% of model releases, report finds
QQQ•A SemiAnalysis review of 857 releases by nine major Chinese AI developers found that 31, or 3.6%, had published safety evaluations tied to a specific model; only nine had results available at or before launch.
1. Limited public disclosure
SemiAnalysis reviewed models released between 2021 and September 15 by Alibaba, ByteDance, Tencent, Baidu, DeepSeek, Moonshot, Z.AI, MiniMax and StepFun. It found no published safety disclosure for 813 releases, while noting that companies could have tested models privately.
2. Rules focus on applications
The report said China’s binding rules principally govern AI applications and their effects on users, rather than requiring frontier developers to publish risk assessments based on model capabilities. It also said no major Chinese developer had released a frontier text model with publicly disclosed dangerous-capability tests spanning cyber, biological and loss-of-control risks.
