Report: AI attacked AI — world’s first automatic cyberattack from OpenAI repelled by Chinese model Zhipu GLM-5

According to a report published on the Hugging Face platform, researchers have recorded the world's first fully autonomous cyberattack carried out by artificial intelligence without human involvement. The agent, based on OpenAI's model, attempted to breach a security system, but its attack was detected and repelled by the Chinese model GLM-5 developed by Zhipu (智谱). The incident was reproduced and analyzed using tools from the Southern China newspaper Southern Metropolis Daily. The experiment demonstrates that AI systems can not only generate content but also autonomously perform malicious actions, raising safety concerns and highlighting the need for defensive AI mechanisms.