China’s Zhipu AI has unveiled its latest model, GLM-5.2, which reportedly matches the performance of leading U.S. AI systems in cybersecurity and software vulnerability detection tasks. This development raises questions about the effectiveness of current U.S. AI export controls.
Released on June 13, 2026, GLM-5.2 is available under an open-weight license, allowing researchers and developers worldwide to access and run the model on standard consumer hardware. In contrast, U.S.-based models like Anthropic’s Claude Mythos are subject to export restrictions, limiting their availability.
While GLM-5.2 may not yet rival U.S. models in general-purpose benchmarks, its targeted performance in vulnerability identification is noteworthy. Independent testing by Semgrep indicates that GLM-5.2 achieved an F1 score of 39% in detecting Insecure Direct Object Reference (IDOR) vulnerabilities, surpassing Claude Code’s 32–37% on similar tasks. Additionally, the cost per vulnerability found using GLM-5.2 is approximately $0.17, significantly lower than the over $1.00 cost associated with Claude-based workflows.
These findings suggest that open-weight models like GLM-5.2 can deliver high performance in specific security domains at a fraction of the cost of proprietary systems. This challenges the assumption that restricting access to advanced AI models would prevent adversaries from developing comparable capabilities.
Anthropic’s Project Glasswing previously demonstrated the potential of AI in vulnerability research by uncovering over 10,000 critical vulnerabilities. The emergence of GLM-5.2 indicates that similar capabilities are now accessible beyond U.S. borders.
As OpenAI introduces GPT-5.6 with limited access due to misuse concerns, the availability of GLM-5.2 underscores the need for a reassessment of AI export control strategies. The rapid advancement of open-weight models in specialized AI domains suggests that hardware restrictions and model access controls may not be sufficient to maintain a competitive edge in AI technology.
The release of GLM-5.2 highlights China’s progress in developing specialized AI applications, prompting a critical evaluation of current policies aimed at preserving technological leadership. The global accessibility of such models necessitates a balanced approach to innovation and security in the AI landscape.
