Taiwan's Ministry of Digital Affairs has officially confirmed a coordinated cyberattack on government institutions carried out using autonomous AI agents. The incident was disclosed on August 13, but according to available data, the attackers had been active since at least July, when national cybersecurity systems recorded the first anomalous activity.
The attack was hybrid in nature: part of the operations was carried out manually, but the main thrust came from the use of AI assistants, including the OpenClaw toolkit. This allowed the attackers to operate with unprecedented speed and scale, unattainable for humans. As a result of the breach, 85 employee accounts were compromised, with 84 of them granting access to internal networks on their own, likely falling victim to phishing or social engineering schemes.
Scale of Damage and Attack Vector
The hackers' loot included more than 2,500 personnel documents, logins and passwords for internal databases, as well as network infrastructure diagrams. The attack was not limited to a single agency—it spread to contractors of government IT systems, including the nuclear safety administration, government email, and seven energy companies. This indicates that the attackers had a clear goal—not just a targeted breach, but deep penetration into the island's critical infrastructure.
Notably, the attack was not technically revolutionary. As experts emphasize, the key role was played not by the novelty of exploits, but by the autonomy of AI agents. They allowed for parallel scanning of numerous systems for vulnerabilities and conducting attacks at machine speed, which fundamentally changes the balance of power in cyberspace. Now, attackers only need typical weaknesses in infrastructure to cause serious damage if they can exploit them automatically.
Statistics confirm a worrying trend: in 2025, the number of cyberattacks on Taiwan's critical infrastructure grew by 6%, reaching an average of 2.63 million incidents per day. Part of this activity is linked by island authorities to "hybrid threats" from China, although direct evidence of a specific state's involvement in this case has not yet been presented.
Analyst Conclusions
This incident is a vivid demonstration that AI agents are becoming the primary tool for a new generation of cybercriminals. Taiwanese authorities managed to contain the threat, but the very fact that the attack lasted for several weeks and affected strategic facilities indicates the need to rethink defense approaches. Traditional perimeter systems can no longer cope with threats where response speed is measured in milliseconds. The market needs proactive AI solutions for defense, not just reactive measures.