China's AI model approaches Anthropic's Mythos 5 in cyber-defense tests

China's AI model approaches Anthropic's Mythos 5 in cyber-defense tests

Qihoo 360's Yitian Tulong (bug hunting: 3,432 flaws) and Zhipu AI's GLM-5.2 narrow gaps with Claude Mythos 5 in cyber-defense. The 20-30% gap between Chinese and US models shrinks through vertical specialization, raising geopolitical cyber-AI risks.