China Reaches Anthropic’s Level in Cybersecurity Capabilities China Has Matched Anthropic in Cybersecurity, Resetting AI Race
The global race for artificial intelligence has entered a new phase following reports that Chinese AI developers have achieved capabilities on par with some of the most advanced US systems in the realm of cybersecurity. This discussion intensified after reports emerged that Chinese AI tools are performing nearly on the same level as Anthropic’s “Mythos”—a cybersecurity-focused AI system that garnered international attention earlier this year.
According to reports and public statements by researchers, China’s progress is not being viewed as overall AI dominance, but rather as a narrowing of the gap in a specific domain: cybersecurity and vulnerability identification. This distinction is significant because cybersecurity AI is increasingly viewed as a strategic capability rather than merely a software product.
Zipu AI (Z.ai) is also emerging in these discussions. Its model, GLM-5.2, has reportedly performed well in specific cybersecurity tasks—particularly in detecting software bugs and conducting security analysis. According to reports, the model’s capabilities in certain benchmark areas were on par with, or close to, those of Anthropic’s cybersecurity-focused systems.
Meanwhile, the Chinese cybersecurity company ‘360 Security Technology’ introduced a distinct set of AI-powered security tools designed as domestic alternatives for advanced cyber operations. The company unveiled systems capable of automatically detecting vulnerabilities and supporting cyber defense workflows. It stated that these systems are part of a broader effort to strengthen China’s cybersecurity ecosystem.
This development has reignited the debate in the US over whether restricting access to domestically developed advanced AI models inadvertently creates opportunities for international competitors. Recent US policies have limited access to certain cutting-edge AI systems due to national security concerns, sparking a discussion on balancing security with the need to maintain technological leadership.
The conversation also encompasses the issue of “open-weight AI models.” Unlike fully closed systems, open-weight models offer developers broader access and the ability to modify the technology. Proponents argue that this accelerates innovation and adoption, while critics warn that the powerful capabilities inherent in open ecosystems could also increase the risk of misuse.
Cybersecurity experts point out that artificial intelligence capable of rapidly identifying software vulnerabilities can prove valuable for both defense and offense. While this technology aids in securing systems, if used carelessly, it can also accelerate the discovery of vulnerabilities. Due to this dual-use nature, governments are increasingly prioritizing advanced artificial intelligence as a form of strategic infrastructure.
The conversation intensified following Anthropic’s allegations that Chinese entities attempted to extract the model’s capabilities on a massive scale through repeated interactions with its systems. While these allegations remain disputed, they have added a new dimension to the broader discourse surrounding AI competition, intellectual property, and model security.
Despite these shifts, analysts maintain that leading US AI labs remain ahead in several key AI categories. However, recent reports suggest that this lead in certain technical areas may not be as substantial as previously thought.
The crucial point is that the AI race is no longer defined by a single revolutionary model; rather, it is about who can successfully integrate research, infrastructure, regulation, security, and large-scale deployment. It is difficult to say whether this moment will truly “reset” the AI race, but it clearly indicates that competition in advanced AI and cybersecurity is intensifying far more rapidly than anticipated.