AI agent swarm used against Taiwanese government systems
What happened
Suspected China-linked operators ran a near-autonomous campaign using open-source agent frameworks, with multiple parallel sub-agents handling reconnaissance, credential attacks and lateral movement. Reported results included 85 cracked accounts and more than 2,500 personnel records taken. Taiwan's Ministry of Digital Affairs confirmed an AI-agent-assisted attack.
Over 12 attack waves between 1 and 4 July, a framework built on the open-source Hermes and OpenClaw agents ran up to eight sub-agents at once, each with its own targets and techniques. Starting from a single government portal, it mapped 21 connected systems from their public web code, found an employee database open without authentication and debug endpoints that handed out logged-in sessions, then used the stolen staff details to guess passwords, solving CAPTCHAs automatically. It went on to collect access secrets and internal database credentials, exploited a flaw in a government authentication service to plant a backdoor, and widened its scanning to the national nuclear safety agency, government IT suppliers, a government email system and more than seven energy companies. The operators got around the agents' safeguards by describing the work as an authorized penetration test.
Israeli firm Dream Security reconstructed the operation from a 160 MB archive of 1,395 files left by the attackers, with internal notes written in Chinese. Taiwan was identified as the target by the Financial Times and The Register.
How it happened
AI-run attack: Autonomous intrusion. The attacker's own AI agent carried out the break-in, working through the steps with little human involvement.
Sources do not say whose AI model was involved. The product was Hermes and OpenClaw agents, which is open source, a type of AI agent.
Impact
Data was exposed. Data involved: personal data and credentials. Many organizations were affected. Sources give a figure of 2,500.
Severity is rated High, calculated from how far the incident reached and how sensitive the data was.
Was the AI at fault?
No. The failure was in the systems, settings or people around the AI, not in the model's behaviour.
Status and timeline
Confirmed by the affected organization, the AI vendor, a regulator or a named security research firm.
It happened or began on 1 Jul 2026. It became public on 12 Aug 2026, 42 days later.
Taiwan's Ministry of Digital Affairs confirmed the attack, saying its investigation found it came from overseas and combined conventional hacking with AI agents such as OpenClaw, but did not attribute it to a specific country.