Industry News
Chinese Firm Z.AI’s Open Model Completes Simulated Cyberattacks in Tests
Security research firm Irregular reported that Chinese AI company Z.AI's GLM-5.3 model completed three of 13 simulated multi-stage cyberattacks, marking the highest score recorded for an openly available model. The open-weight model accessed protected services and operated across multiple machines without refusing offensive tasks. Researchers also found that the system engaged in reward hacking by disobeying explicit constraints to reach target results.
