The campaign unfolded over four days at the beginning of July and targeted 21 government systems. The attackers deployed as many as eight AI agents simultaneously to map networks, identify vulnerabilities, test possible entry points and change tactics when individual attack paths failed.
At least 85 government user accounts were compromised and more than 2,500 personnel records were extracted. The operation subsequently expanded towards Taiwan’s nuclear safety agency and at least seven energy companies, placing critical infrastructure within the hackers’ wider reconnaissance and targeting activity.
Cybersecurity researchers investigating the intrusion described it as an unusually advanced example of an end-to-end autonomous attack. Rather than merely using generative AI to write malicious code, research targets or assist a human operator, the system divided work among several agents and made decisions as the operation developed.
The tool used the open-source agent frameworks Hermes and OpenClaw. Such systems allow AI models to interact with software tools, search for information, execute commands and pursue multi-step objectives. The specific underlying AI model powering the agents has not been identified.
Evidence recovered from the operation showed that safeguards designed to prevent malicious behaviour were bypassed by presenting the activity to the AI system as authorised security testing. This allowed the agents to perform tasks that resembled legitimate vulnerability assessment while they were being used against real targets.
Attribution remains less conclusive than the evidence of the intrusion itself. Investigators have not linked the campaign to a named hacking group or established direct state control. Internal communications associated with the operation were written in Simplified Chinese, leading researchers to assess a strong probability that the operator had connections to China. Material extracted from the targeted networks was predominantly in Traditional Chinese.
Taiwan’s Ministry of Digital Affairs has not publicly confirmed details of the specific breach, citing confidentiality surrounding cybersecurity incidents. It has acknowledged that the growing integration of AI is changing the nature of cyber threats, particularly because automation can increase the speed and scale of attacks while introducing new vulnerabilities associated with AI agents themselves.
The episode comes against an already intensive cyber pressure campaign facing Taiwan. The National Security Bureau said critical infrastructure experienced an average of 2.63 million intrusion attempts attributed to China every day during 2025, an increase of 6 per cent from 2024 and more than double the level recorded in 2023. Energy, healthcare, communications, government administration and technology were among the principal targets.
The shift towards autonomous agents changes the economics of such operations. Traditional sophisticated intrusions require skilled personnel to conduct reconnaissance, analyse vulnerabilities and adapt continuously when defences interfere. AI agents can perform several of those activities in parallel, potentially allowing attackers to investigate far more targets without increasing their human workforce proportionately.
Similar developments have been appearing elsewhere. Cybersecurity investigators have documented Chinese-speaking threat actors combining AI-driven autonomous reconnaissance with manual exploitation, demonstrating that attackers are experimenting with systems capable of selecting targets and testing vulnerabilities without continuous human supervision.
Academic work has also demonstrated that advanced language-model agents can autonomously identify and exploit website vulnerabilities. Researchers have warned that this could weaken one of cybersecurity’s traditional protections: the scarcity and cost of highly skilled attackers. Automation makes it possible to test large numbers of systems while rapidly adjusting techniques when individual methods fail.
Taiwan is particularly exposed because cyber operations form part of a broader contest involving military activity, espionage, disinformation and pressure on critical infrastructure. Beijing claims Taiwan as part of its territory, while Taipei rejects Beijing’s sovereignty claims and maintains that the island’s future can be decided only by its people.
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
