In recent years, the competition in artificial intelligence (AI) between China and the United States (US) has become increasingly fierce. A notable development in this rivalry is the introduction of Z.ai's latest AI model, GLM-5.2, which boasts impressive capabilities in detecting software vulnerabilities, comparable to the Claude Mythos model by Anthropic.
GLM-5.2's Cybersecurity Capabilities
According to a report from The Wall Street Journal, security researchers have tested GLM-5.2 and found it effective at identifying security vulnerabilities in software, matching the performance of Claude Mythos. This is a significant advancement, as more companies are turning to AI technologies to identify and patch security gaps before they can be exploited by cybercriminals.
Open-Source Advantage and Concerns
One of the standout features of GLM-5.2 is its open-source nature, allowing it to be downloaded, modified, and operated independently without relying on cloud services from AI providers. This characteristic makes it particularly attractive for businesses looking for greater flexibility in developing cybersecurity solutions.
However, the open-source aspect also raises concerns. The availability of an easily accessible AI model might enable cybercriminals to design or refine attacks against computer systems. Despite these worries, the report emphasizes that while GLM-5.2 is advancing in cybersecurity, it does not mean that China has overtaken the US in overall AI development.
Performance Comparison with US Models
In various general tasks, GLM-5.2 still lags behind AI models from Anthropic and OpenAI. Nevertheless, the performance gap in cybersecurity between Chinese AI and leading US models has reportedly narrowed significantly. Benchmark data cited in the report indicates that GLM-5.2 even outperformed Claude Opus 4.8 in several security tests.
Researchers noted that GLM-5.2's capabilities could further improve to match Claude Mythos in bug detection when combined with appropriate prompting techniques.
The Intensifying AI Race
The launch of GLM-5.2 signals a tightening AI race between China and the US. Recently, US AI firms like Anthropic and OpenAI have started to restrict access to their most advanced AI models, citing national security concerns. In contrast, many AI laboratories in China are increasingly releasing open-weight AI models that the public can download and use.
This trend corresponds with ongoing debates regarding how swiftly China can catch up to American AI capabilities. Elon Musk had previously predicted that Chinese AI laboratories would not match the benchmarks of Anthropic's flagship model, Fable 5, until the first quarter of 2027. However, Tang Jie, the founder of Zhipu AI, refuted this claim, expressing confidence that the gap would close much sooner.
Source: https://tekno.kompas.com/read/2026/07/06/19090077/bukti-kecanggihan-ai-china-makin-dekati-as
