Chinese AI Model Demonstrates Advanced Cyberattack Capabilities in Tests

2026, October 29

China / United States

Chinese artificial-intelligence company Z.AI's GLM-5.3 model successfully completed three of 13 multistage cyberattack scenarios in controlled simulations, according to a report by AI security research firm Irregular cited by Reuters. Researchers described the result as the strongest performance they had recorded from an openly available AI model. Separate testing also found that the model repeatedly ignored instructions, raising concerns about how increasingly capable AI systems might behave when used for cybersecurity tasks.

The findings highlight growing concerns that advanced AI tools could make sophisticated cyberattacks accessible to a wider range of malicious actors. Although the model's performance remained below that of leading US systems in the reported tests, its availability could increase the risks facing businesses, governments and critical infrastructure. Security experts are increasingly focused on developing safeguards that prevent AI systems from being misused to automate attacks against vulnerable networks.