# Chinese AI matches Anthropic in cybersecurity capabilities, raising alarms over US export controls > Source: > Published: 2026-06-29 17:05:00+00:00 # Chinese AI matches Anthropic in cybersecurity capabilities, raising alarms over US export controls Zhipu AI's GLM-5.2 rivals Anthropic's Mythos in bug-finding and vulnerability detection, while costing less to run The cybersecurity AI arms race just got a new front-runner, and it’s not from San Francisco. Zhipu AI’s GLM-5.2, an open-weight model released in June 2026, has matched Anthropic’s Mythos model in key cybersecurity tasks like bug-finding and vulnerability detection, according to benchmarks conducted by Semgrep and reported by The Wall Street Journal. GLM-5.2 actually outperformed Anthropic’s Claude Opus 4.8 in certain tests. And it did so while being more cost-effective. ## The benchmarks tell the story Semgrep, the code analysis platform, ran GLM-5.2 through cybersecurity-specific benchmarks that tested its ability to find bugs and detect vulnerabilities. The Chinese model performed on par with Mythos, Anthropic’s flagship cybersecurity-capable system, across these tasks. GLM-5.2 now ranks among the top ten most used AI models on OpenRouter. 360 Security Technology, a major Chinese cybersecurity firm, unveiled two new AI-based systems on June 24, 2026. The systems, called “Tulongfeng” and “Yitianzhen,” are being positioned explicitly as China’s answer to Mythos. ## Why open-weight models change the calculus GLM-5.2 is an open-weight model. Anyone can download it, run it, and modify it without asking Zhipu AI for permission. This is fundamentally different from how Anthropic or OpenAI operate, where access flows through APIs with usage policies and safety guardrails baked in. A model that excels at finding vulnerabilities in code is, by definition, a model that could help attackers find vulnerabilities in code. When that model sits behind an API with terms of service and monitoring, there’s at least a theoretical layer of control. When it’s open-weight and downloadable, that layer evaporates. Industry experts quoted in WSJ reporting have flagged this exact concern. The proliferation risk isn’t hypothetical. It’s a direct consequence of the open-weight architecture that makes these models so attractive and accessible in the first place. ## The export control paradox The US has spent the last several years tightening export controls on advanced chips and AI technology, specifically to slow China’s AI development. GLM-5.2 suggests that strategy has, at minimum, not worked as intended. Some analysts are now arguing that US restrictions may have inadvertently accelerated China’s push toward self-sufficiency in AI. The backdrop here is US-China tensions over AI that extend well beyond commercial competition. State-linked actors have been increasingly active in cyber operations, and the availability of frontier cybersecurity AI models, particularly open-weight ones with no usage restrictions, adds a new variable to an already volatile equation. **Disclosure:** This article was edited by Editorial Team. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).