Chinese AI Matches Anthropic in Cybersecurity, Reshaping Global Competition

U.S. companies Anthropic and OpenAI have long dominated frontier AI, with cybersecurity capabilities seen as critical to national security and corporate defenses. Zhipu AI, also known as Z.ai, is now narrowing the gap with downloadable, locally deployable open-weight models. The technology lowers the cost of vulnerability scanning but also makes it easier for attackers to remove safety restrictions, forcing companies to patch systems faster and increasing regulatory pressure.

On June 13, 2026, Zhipu AI released GLM-5.2, which uses a mixture-of-experts architecture. Semgrep tests showed that the model achieved an F1 score of 39% on some vulnerability-detection tasks and cost about $0.17 for each vulnerability found, outperforming Claude Opus 4.8 and GPT-5.5. In late June, 360 Security Technology separately said its “Tulongfeng” system had identified more than 3,400 vulnerabilities.