Taiwan has become the first country to confirm an attack carried out by autonomous AI agents, after its national cybersecurity agency disclosed an "unusual" intrusion that took place in July. According to a report by Israeli cybersecurity firm Dream, attackers deployed open-source AI agents named Hermes and Openclaw across twelve waves of assault, with up to eight agents operating simultaneously — stealing roughly 1,400 documents, compromising credentials for 85 government employee accounts, and installing backdoors in government web applications, all within four days. Dream suspects Chinese-state-linked hackers based on linguistic analysis showing the use of simplified Chinese characters in operational documentation, and experts warn the incident signals a dangerous new era: AI systems that continuously adapt their attack strategies in real time give hackers a growing asymmetric advantage over defenders, who must protect against every possible vulnerability while attackers need find only one.