Z.ai releases GLM-5.2 and researchers flag cybersecurity strength

In short: China’s Z.ai has released GLM-5.2, and researchers say it can perform like Anthropic’s Mythos on some cybersecurity tasks.

What happened

China’s Zhipu AI, which also uses the name Z.ai, released a new AI model called GLM-5.2. Some researchers say GLM-5.2 can match Anthropic’s Mythos in certain bug-finding and cybersecurity tests.

A “bug” is a mistake in software code, and some bugs become security holes that hackers can use. These AI models can act like a fast assistant for security work, scanning code and suggesting where problems might be.

The Verge reports that GLM-5.2 still trails leading US models from Anthropic and OpenAI on more general tasks. But it appears to have narrowed the gap on security focused work, especially finding vulnerabilities (weak spots in software, like an unlocked window in a house).

Another key detail is that GLM-5.2 is “open-weight.” That means the core model files can be downloaded and run by other people, instead of only being available through a company controlled service.

Why it matters

US officials have been trying to limit China’s access to top AI models and the computer chips used to train and run them. The concern is that AI that is good at finding security holes can also be used to break into systems, and open-weight models are harder to monitor because anyone can run them with little oversight.

Source: The Verge AI