
Suspected China linked operators built an autonomous attack platform out of publicly available AI agent frameworks, Hermes and OpenClaw, and pointed it at Taiwan's government. Over four days in early July the system ran as many as eight agents at once. The agents mapped 21 government systems, researched vulnerabilities, changed tactics when they were blocked, and moved through the network with minimal human direction.
The reported result is at least 85 accounts and 2,500 personnel records compromised, with the operation spreading to Taiwan's nuclear safety agency and at least seven energy companies. The Israeli firm Dream, which uncovered the campaign, describes it as the first fully autonomous attack of its kind on a foreign government. Attribution rests on operational artefacts, including internal communications in Simplified Chinese against data pulled in Traditional Chinese.
Nothing here required a frontier lab or a state budget. The frameworks are open source. The significant change is not capability, it is cost: continuous adaptive attack that previously needed a skilled team working shifts now needs a machine and someone to point it.
Defences built on the assumption that attackers get tired, work business hours, or run out of patience are defences built on an assumption that has expired. Incident readiness under AISOF organisational requirements assumes pressure that does not stop, and asks what is detected, by whom, and how fast, rather than what the perimeter is supposed to prevent.
Sources
Briefings summarise public reporting for educational purposes and are not legal advice. Editorial policy