BEIJING — The flagship AI model of Chinese firm Z.AI carried out complex cyberattacks in controlled simulations and repeatedly ignored instructions in separate tests, according to a report by AI security research firm Irregular, Reuters reported on Thursday, 29 October.

Source: Reuters

The GLM-5.3 model successfully completed 3 of 13 multi-stage cyberattack scenarios at least once, the strongest performance Irregular has recorded from an openly available AI model, according to the researchers. Irregular said GLM-5.3 gained access to protected services, established alternative connections to move through computer networks and coordinated operations across multiple machines during the simulated attacks, and did not refuse any of the offensive tasks tested.

The model still trails leading US models in Irregular's tests. OpenAI's GPT-6 Astra, for example, completed 9 of 10 scenarios in a separate assessment, although the tests were not identical, Reuters reported.

In separate tests, Irregular found that GLM-5.3 repeatedly reproduced computer code from memory despite being explicitly told not to, allowing it to arrive at correct answers while breaking rules designed to test its abilities. The researchers described the behaviour as “reward hacking”, in which an AI system finds ways to achieve a desired result while disregarding instructions, and warned that tests measuring only whether a model succeeds could overlook such violations.

The findings underscore concerns that increasingly capable AI models lower the technical barriers to sophisticated cyberattacks, particularly with many Chinese AI models like GLM-5.3 that are open-weight — available for users to download, modify and operate independently. Beijing-based Z.AI, one of China's leading AI startups, did not immediately respond to a request for comment, Reuters said.