Four days, 21 systems mapped: Taiwan describes the first fully autonomous cyber attack against a state

Over four days in July, a system coordinating up to eight AI officers mapped 21 Taiwanese government systems, broke 85 accounts and extracted 2,500 personnel files without human intervention.

Taiwanese authorities have described what experts say is the first known case of a fully autonomous attack on government agencies. Over four days in July, deployed AI agents mapped 21 state systems, forced 85 user accounts and exfiltrated 2,500 personnel records, according to Israel's Dream, which detected the intrusion. Taiwan's Ministry of Digital Affairs cites clear indications of a foreign origin and a hybrid approach mixing conventional operations and agents like OpenClaw; neither Taiwan nor Dream has publicly confirmed the origin of the attacks, which experts suspect to be Chinese.

The difference with regular campaigns is the degree of autonomy. Using a model to write code or look for vulnerabilities has become commonplace; here, a system coordinated up to eight agents who carried out reconnaissance, credential attacks, and the choice of subsequent attack paths, deciding for themselves what to do next. For Kenny Huang, president of the Taiwan Network Information Center, this is the first disclosed case of a fully automated attack targeting a government.

Amir Becker, head of strategy at Dream, underlines the ability of the system to adapt: when an approach was blocked, the system looked for new techniques in real time and adjusted its tactics. The company sums up the defensive challenge in one sentence: the cost of conducting a competent attack has collapsed, the cost of defending against it has not.