Chinese AI Model Challenges Anthropic in Cyber Tests

Chinese AI startup Z.ai has unveiled claims that its latest open-source GLM-5.3 model is closing the gap with Anthropic’s restricted Mythos 5 in cybersecurity capabilities, though it remains behind in developing working exploits.

Z.ai said GLM-5.3 achieved an 84.5 per cent score in the CyberGym vulnerability-detection test, compared with 83.8 per cent for Mythos 5. However, its ExploitBench score was 54.4 per cent, well below Mythos 5’s 78 per cent.

The results have not been independently verified.

Z.ai plans to publicly release GLM-5.3 after completing security assessments, while restricting its most sensitive cybersecurity features to verified users. The company says additional safeguards will be used to prevent malicious applications.

The launch highlights intensifying competition in advanced AI development and growing debate over the risks and benefits of releasing powerful cybersecurity models as open-source systems.