NewsBite analysis · AI, Security, Startups · New Product, Open Source
The Economic Times
China's Z.ai says new model nears Anthropic's Mythos 5 in cyber-defence tests
Z.ai, a Chinese AI startup, claims its GLM-5.3 model has nearly matched Anthropic's Mythos 5 in a series of cybersecurity tests, showing notable improvements particularly in identifying software vulnerabilities. While GLM-5.3 scored 84.5% on CyberGym, slightly above Mythos 5's 83.8%, it lagged in exploiting these vulnerabilities. This announcement underscores the rising capabilities of Chinese AI models in a competitive space typically dominated by Western firms. Z.ai aims to release GLM-5.3 publicly, ensuring a controlled access to its more advanced security features akin to Mythos 5's vetted deployment. This approach reflects a broader trend of increased sophistication in safety considerations within China's AI landscape.
MonitorConfidence: MediumShort term horizon
Published 14 Aug 2026, 14:30 UTC5 min read
Visit original article ↗Lead image — The Economic Times
Summary
TL;DR
- Z.ai's GLM-5.3 scores 84.5% in vulnerability detection tests.
- Model performance closely rivals Anthropic's Mythos 5 in cybersecurity.
- Z.ai plans a public release with stringent access for sensitive features.
- Competitors like Hugging Face have reported successful implementations of prior Z.ai models.
- Critics raise concerns over safeguarding of open-source AI models.
Interesting facts
- This marks a significant development for Chinese AI in cybersecurity capabilities.
- GLM-5.3's training methodology includes extended task environments.