Zhipu AI Matches Claude in Vulnerability Detection

China's Zhipu AI has released a new model that reportedly matches the performance of Anthropic's Claude Mythos in detecting software vulnerabilities. The GLM-5.2 model is open-weight and freely accessible worldwide. What's interesting is that GLM-5.2 achieved an F1 score of 39% for IDOR vulnerability detection, surpassing Claude Code's 32–37%. The cost to find each vulnerability is also significantly lower, at approximately 17 cents compared to over one dollar for Claude-based workflows. This development challenges U.S. export controls aimed at preventing foreign entities, including Chinese researchers, from accessing advanced AI models. It raises concerns that powerful AI capabilities are no longer exclusively held by the U.S. and could accelerate both defensive automation and potential cyber threats globally.