AI agents breach Taiwan government networks

Autonomous artificial intelligence agents linked to China compromised Taiwan government systems, marking a significant escalation in AI-assisted cyber espionage. Over four days in early July, the campaign targeted 21 government systems, deploying up to eight AI agents simultaneously to map networks, identify vulnerabilities, and adapt tactics when necessary.

At least 85 government user accounts were compromised, with over 2,500 personnel records extracted. The intrusion expanded to Taiwan’s nuclear safety agency and at least seven energy companies, posing risks to critical infrastructure. Researchers described the operation as an unusually advanced example of an end-to-end autonomous attack, using open-source agent frameworks Hermes and OpenClaw to divide tasks among agents and make decisions throughout the operation.

Evidence suggests safeguards were bypassed by presenting the activity as authorized security testing, allowing the agents to perform tasks resembling legitimate vulnerability assessment while targeting real systems. Attribution remains unclear, with investigators unable to link the campaign to a specific hacking group or confirm direct state control.

Internal communications were written in Simplified Chinese, indicating a strong probability of Chinese connections. The targeted networks produced material predominantly in Traditional Chinese. Taiwan’s Ministry of Digital Affairs has not publicly confirmed the breach, citing confidentiality concerns. The incident comes amid an intense cyber pressure campaign targeting Taiwan, with an average of 2.63 million intrusion attempts attributed to China daily in 2025, a 6% increase from 2024 and more than double 2023 levels.

The use of autonomous AI agents changes the economics of cyber operations, allowing attackers to investigate more targets without increasing their human workforce proportionately. This development aligns with similar trends observed elsewhere, where Chinese-speaking threat actors combine AI-driven autonomous reconnaissance with manual exploitation.

Academic research also demonstrates that advanced language-model agents can autonomously identify and exploit website vulnerabilities, potentially weakening cybersecurity protections by reducing the scarcity and cost of highly skilled attackers. The article AI agents breach Taiwan government networks appeared first on Arabian Post.