Taiwan Government Confirms AI-Driven Cyberattack on State Agencies

Taiwan's Ministry of Digital Affairs (MODA) and the Administration for Cyber Security have confirmed that government agencies were targeted in a cyberattack last month involving artificial intelligence (AI) agents. The intrusion, first detected by the Israeli AI startup Dream, utilized open-source AI agents such as OpenClaw to build autonomous hacking tools. These tools compromised at least 85 government user accounts and extracted thousands of personnel records. The attackers also targeted Taiwan's nuclear safety agency and at least seven energy companies.

Technical analysis from Dream indicates that the AI agents operated with a high degree of autonomy. The system reportedly evaluated attack paths based on success probability and adjusted its strategy in real time. When one path was blocked, the system would deploy another agent to research new methods. The attack involved twelve waves of activity, with up to eight AI agents operating in parallel.

While the Taiwanese government stated the attacks originated overseas, it did not officially name a specific country. However, researchers at Dream suggested the operators were likely connected to China because internal communications were written in simplified Chinese. A center-left source noted that the National Security Bureau previously reported a rise in Chinese cyberattacks on key infrastructure, often synchronized with military drills.

Perspectives on the autonomy of the attack vary. A center-right source framed the event as a landmark moment in cyber warfare, suggesting it was the first time a country was attacked with autonomous AI agents. Conversely, a center-left source included expert caution, noting that a human operator must still establish objectives and directives, meaning the process is not entirely autonomous.