Researchers Report Near-Autonomous AI Attack on Taiwan Government
Summary
Dream researchers say suspected Chinese hackers used Hermes and OpenClaw in a near-autonomous operation against Taiwan’s government, reportedly extracting over 2,500 personnel records and expanding to government suppliers and other organizations.
Key points
- Dream researchers attributed the operation to suspected Chinese hackers; the article does not independently verify that attribution.
- Dream said the operation extracted more than 2,500 personnel records and achieved real-world compromises.
- The operation reportedly expanded to government IT vendors, a nuclear safety agency, a government email system and seven or more energy companies.
- The framework used the open-source AI frameworks Hermes and OpenClaw, with autonomous “Learning Cycles” to research target-specific techniques and adapt during the operation.
- Dream said the attackers framed their work as authorized penetration testing to bypass AI safety guardrails.
- The researchers noted that the system still required human adjustment, including tuning agent coordination and decision logic.
Article Details
- Event Type
- Reported near-autonomous AI-enabled cyberattack against the Taiwanese government
- Impact
- Researchers at Dream reported that the operation compromised state infrastructure and extracted more than 2,500 personnel records. It also expanded to government IT vendors, a nuclear safety agency, a government email system, and energy companies.
MITRE ATT&CK
Tools
HermesSuspected Chinese hackers used open-source AI frameworks Hermes and OpenClaw to launch a near-autonomous cyberattack against the Taiwanese government, marking the first publicly known AI-driven attack to hit aOpenClawSuspected Chinese hackers used open-source AI frameworks Hermes and OpenClaw to launch a near-autonomous cyberattack against the Taiwanese government, marking the first publicly known AI-driven attack to hit a
Countries
Industries
Energy2,500 personnel records and expanded to vendors, a nuclear safety agency, a government email system, and energy companies while adapting through “Learning Cycles” without human intervention.GovernmentAI frameworks Hermes and OpenClaw to launch a near-autonomous cyberattack against the Taiwanese government, marking the first publicly known AI-driven attack to hit a government target.