The International AI Security Rankings were announced today: Chinas open-source solution ranks third globally and first in open source.

Today, the internationally recognized AI security benchmark CyberGym announced its latest rankings. DoGNAVY, jointly developed by China's top artificial intelligence research institutions and the security team DaKunNuoWei, ranks third in the world with a pass rate of 90.8%, and first among open source solutions. CyberGym is currently the largest network security testing platform for AI agents. How did the Chinese team manage to break into the top three? Two technical pathways provide the answers. In the lead are Microsoft and Google, which employ multiple top-tier closed-source model combinations for their operations, routing each segment to the best-performing modelprovided that one can afford and access the world's strongest closed-source models simultaneously. DoGNAVY, on the other hand, takes a different approach: it uses just one open-source general model that anyone can download and deploy freely. Notably, the Chinese solution DoGNAVY utilizes only one open-source general model, while its higher-ranking competitors rely on combinations of multiple closed-source models. AI security offense and defense are entering a practical stage, and the open-source route is enabling more innovators to acquire top-notch security capabilities.