At ISC.AI 2026 in Beijing on June 24, Qihoo 360 founder Zhou Hongyi unveiled Tulong Feng, an AI vulnerability agent positioned as China's answer to Anthropic's Mythos, alongside the Yitian Zhen automated defense platform and the Panshi Zhidun security coalition. Zhou criticized the U.S. export controls on Mythos, which restrict Chinese access via the Glasswing program, calling such AI systems 'cyber nuclear weapons.' He claimed Tulong Feng has discovered 3,432 vulnerabilities, with 105 confirmed by Chinese regulators. Meanwhile, Beijing-based Z.ai (Zhipu AI) released GLM-5.2 under an MIT license shortly after Anthropic's models went offline for foreign nationals. GLM-5.2 scored 39% on Semgrep's insecure direct object reference detection test (F1 metric), ahead of Claude Code, and matched Claude Opus 4.8 on a Graphistry capture-the-flag challenge at roughly $0.17 per finding, compared to over $1 for Claude workflows. Z.ai co-founder Tang Jie called Anthropic's withdrawal regrettable, while technical lead Qinkai Zheng emphasized open accessibility. Elon Musk predicted China wouldn't match Fable-level capability until Q1 2027, but Tang responded: 'Won't take that long.'

Key facts

• Qihoo 360's Tulong Feng claims 3,432 vulnerabilities found, 105 confirmed by Chinese regulators.
• Z.ai's open-source GLM-5.2 scored 39% on Semgrep's IDOR detection test, beating Claude Code.
• GLM-5.2 costs $0.17 per vulnerability finding vs. $1+ for Claude workflows.
• China criticizes U.S. export controls on Anthropic's Mythos, calling it 'cyber nuclear weapons.'
• Z.ai co-founder Tang Jie responds to Musk's prediction: 'Won't take that long.'

KeyAudit data perspective