Z.ai's Open-Source Model GLM-5.3 Challenges Cybersecurity Norms

Chinese AI startup Z.ai's open-source GLM-5.3 model challenges the restricted Mythos 5 in identifying software vulnerabilities. Although it scores slightly higher in identifying security flaws, it lags in attack development. GLM-5.3 aims to democratize advanced cyber-defence tools, advocating for broader access to cybersecurity capabilities.

Chinese AI startup, Z.ai, announced that its open-source GLM-5.3 model nearly matches Anthropic's restricted Mythos 5 in identifying software vulnerabilities, marking significant progress for the Chinese AI sector.

The GLM-5.3 scored 84.5% on the CyberGym test, which evaluates a model’s ability to review code and identify security flaws. This performance slightly surpassed Mythos 5’s 83.8% score. However, in task conversions, GLM-5.3 fell short of Mythos 5, indicating room for improvement in certain areas of defensive security research.

Z.ai plans to publicly release the model in two weeks, with security features accessible through a 'trusted access' programme. The release critiques restricted-access models like Mythos, emphasizing the need for open-source tools to empower smaller security teams globally.