- The attack lasted four days and involved up to eight AI agents operating simultaneously.
- The attackers compromised at least 85 government accounts.
- Researchers consider the incident one of the first examples of a fully autonomous cyberattack.
Chinese hackers used publicly available artificial intelligence tools to attack Taiwan’s government systems. Financial Times reported this, citing researchers at Israeli company Dream, who found traces of the operation.
The statement said the attack lasted four days in early July. The attackers used up to eight AI agents at the same time, which independently mapped the network, searched for vulnerabilities, and changed tactics after failed attempts.
Overall, the tool probed 21 government systems, compromised at least 85 accounts, and obtained more than 2,500 employee records. The attack then spread to Taiwan’s Nuclear Safety Agency and at least seven energy companies.
The Attack Could Have Been Fully Autonomous
According to Dream, the tool used two open AI agent systems — Hermes and OpenClaw. They allow AI models to carry out tasks autonomously.
When one attack method didn’t work, the system brought in another agent. It searched for additional information online and developed a new way to gain access. Researchers also found that the base AI model’s safety guardrails could be bypassed by presenting hacking activity as authorized vulnerability testing.
At the same time, internal messages linked to the attack used simplified Chinese. Researchers believe this points to a high likelihood that the operators were connected to China. Dream did not attribute the attack to any specific group.
A spokesperson for Taiwan’s Ministry of Digital Affairs did not confirm the specific breach, citing confidentiality.
At the same time, he noted that the use of AI is changing the nature of cyberattacks. According to him, AI agents create a double challenge: attacks become automated, and the agents themselves can turn into new vulnerabilities.
Dream researchers called the July incident a new stage in the evolution of cyberattacks. In their view, governments should now assume they may be under constant attack, with a significant share of the work carried out autonomously by AI systems.
Сообщение China-Linked Hackers Launch Autonomous AI Attack on Taiwan Government Systems появились сначала на INCRYPTED.