AI agents built on open-source frameworks conduct multi-day cyberattack on Taiwan nuclear safety agency

According to cybersecurity firm Dream, multiple AI agents constructed from open-source Hermes and OpenClaw frameworks breached Taiwanese government systems over four days in early July, compromising 85 government user accounts, extracting 2,500+ personnel records, generating 1,395 files, and probing the nuclear safety agency before establishing persistent access. Researchers said the attack operated in near-autonomous parallel, with up to eight sub-agents assigned to separate targets, signaling what Dream characterized as a new phase in AI-enabled operations.