Chinese company Zhipu has launched a new AI model, GLM-5.3, which it claims possesses bug-finding capabilities comparable to leading American models. The model demonstrated strong performance on the CyberGym benchmark, a test designed to assess AI's ability to solve real-world cybersecurity challenges, according to a recent report by The Register.
GLM-5.3 reportedly outperforms models like Fable 5 and GPT-5.6 Sol on the CyberGym benchmark, showing an advanced ability to identify vulnerabilities and even construct exploitation chains. Zhipu reported testing the model on real-world codebases, discovering 2,436 vulnerabilities across 269 projects, including 1,097 medium-to-high severity issues. These vulnerabilities spanned various areas, including system kernels, operating systems, browser engines, and network protocols, with some dating back approximately 40 years.
While GLM-5.3 performed less effectively on other security and coding benchmarks compared to Western models, its rapid development in vulnerability discovery suggests China may be closing the gap in AI-driven cybersecurity capabilities.
Source: The Register
