China's Zhipu AI Says Its Model Matches US Anthropic Mythos as a ‘Cyber Nuclear Weapon’

Chinese artificial‑intelligence firm Zhipu AI (Z.ai) announced that its GLM‑5.2 model can locate software vulnerabilities with performance comparable to Anthropic’s Mythos model, while costing roughly one‑sixth as much. The company called the capability a “cyber nuclear weapon,” warning that such power could shift the balance in cyber‑attack and defence. Founder Zhou Hongyi told a security conference that “a weapon of this strength that can change the balance of cyber attack and defence cannot stay only in the hands of others.” Anthropic’s Mythos, which identified critical bugs across major operating systems and browsers, had previously been restricted by the US Commerce Department before limited re‑deployment to trusted entities. Chinese cybersecurity firm 360 Security also claims to have rival models. Experts say the assertions lack independent verification and may serve as deterrent rhetoric, but the development narrows the gap between Chinese and US AI capabilities in a strategic security domain and could pressure US policy on AI regulation.