Zhipu AI Matches Claude in Vulnerability Detection

A new report reveals that China's Zhipu AI GLM-5.2 model is performing on par with Anthropic's Claude Mythos in certain cybersecurity vulnerability detection tasks. This development is intensifying concerns within the U.S. government regarding its AI export control strategy. Here's the thing: Zhipu AI released GLM-5.2 under an open-weight license, making it freely accessible worldwide. This is unlike Anthropic's Mythos, which is subject to U.S. export controls. What's interesting is that GLM-5.2 achieved an IDOR vulnerability detection F1 score of 39%, surpassing Claude Code's 32–37%. It did this at approximately $0.17 per vulnerability found, which is roughly one-sixth the cost of comparable Claude-based workflows. The bottom line is that a freely downloadable Chinese open-weight model can now match U.S. frontier AI in specific security domains, challenging the effectiveness of current export controls.