China's Zhipu AI releases GLM-5.2, matching Anthropic Mythos on cybersecurity benchmarks
Tags AI · Security · Defense Tech
Zhipu AI (Z.ai) released GLM-5.2, an open-weight model that Semgrep's cybersecurity benchmarks show performing on par with Anthropic's Mythos in bug-finding and vulnerability discovery tasks. While the model trails US frontier models on general reasoning benchmarks, the cybersecurity parity represents a significant narrowing of the gap between Chinese and American AI capabilities. The Trump administration has sought to restrict China's access to advanced models like Mythos, viewing them as national security threats. Because GLM-5.2 is open-weight, it can be downloaded and run on commodity hardware by anyone.
Technical significance
The open-weight nature of GLM-5.2 means advanced cybersecurity AI capabilities are now accessible without API gatekeeping, lowering the barrier for both legitimate security researchers and malicious actors. This compresses the timeline for defenders to adopt AI-assisted vulnerability scanning while simultaneously enabling offensive use cases that were previously limited to those with access to closed US models.