On August 13, Taiwan's Ministry of Digital Affairs revealed details of a large-scale cyberattack on government institutions, during which attackers actively used AI agents. The incident, first recorded back in July, has become an alarming signal for the entire region, demonstrating how autonomous algorithms are radically changing the landscape of digital threats.
According to my data, the attack was external in nature and carried out in a combined manner: both manually and through artificial intelligence-based assistants, including the OpenClaw toolkit. The National Institute of Cyber Security detected anomalous activity in government systems, after which agencies received warnings, and affected organizations took necessary protective measures. All systems are currently operating normally, but authorities have already developed new monitoring and protection protocols against such threats.
Attack Mechanics: Speed and Autonomy
An analysis of the incident, which I conducted based on open data, shows the alarming effectiveness of AI tools. Israeli cybersecurity specialists reconstructed the attackers' working environment. Over four days, the system carried out 12 waves of attacks using publicly available utilities. The result — 85 compromised employee accounts, with 84 of them hacked or opened by the users themselves. The attackers' loot included more than 2,500 HR documents, logins, passwords, and network infrastructure diagrams.
Of particular concern is that the attack quickly spread to contractors of government IT systems, the nuclear safety agency, government email, and seven energy companies. The key role here was played not by technical novelty, but precisely by the autonomy of AI. These agents are capable of simultaneously scanning multiple systems for vulnerabilities and exploiting them at machine speed, which is unattainable for humans.
Notably, in 2025, the number of cyberattacks on Taiwan's critical infrastructure increased by 6%, reaching an average of 2.63 million per day. The island's authorities attribute part of this activity to "hybrid threats" from mainland China.
Earlier, CertiK analysts had already warned about the risks of OpenClaw, including data leaks and prompt injections. Now we see that these warnings were not theoretical — AI agents are becoming real weapons, and the balance of power in cyberspace is shifting in favor of attackers. My conclusion: government structures and businesses urgently need to rethink their defense strategies, betting on behavioral analysis and adaptive response systems rather than static security perimeters.