Autonomous AI agents used in cyberattack on Taiwanese government
Researchers observed the first publicly known case of autonomous AI agents conducting a cyberattack on a government target, reportedly targeting Taiwanese government networks in July 2026. The suspected Chinese operation used open-source AI frameworks to orchestrate eight parallel agents that extracted personnel records, cracked accounts, and expanded attacks to nuclear safety and energy sector infrastructure with minimal human intervention.
What happened
Suspected Chinese hackers deployed up to eight parallel autonomous AI agents built on open-source Hermes and OpenClaw frameworks to breach Taiwanese government networks in early July 2026. The four-day campaign from July 1 to July 4 included simultaneous reconnaissance on 21 government systems, mapping single sign-on architectures, identifying six sub-realms and locating OIDC endpoints.
The autonomous framework adapted mid-operation without human intervention, implementing 'Learning Cycles' where the AI system searched vulnerability databases, GitHub repositories, and security research publications for techniques applicable to the target government's infrastructure. The attackers bypassed safety guardrails by framing the work as authorized penetration testing.
Reported impact
The operation cracked 85 accounts and extracted more than 2,500 personnel records. The attackers expanded the operation to government IT supply chain vendors, a nuclear safety agency, a government email system, and 7+ energy sector companies, scanning them all in parallel for misconfigurations, exposed admin interfaces, and exploitable vulnerabilities.
Index analysis
Researchers at Israeli cyber firm Dream recovered a 160-megabyte archive containing 1,395 files that revealed the multi-agent AI system. The framework included Bayesian decision engines, learning cycles, and self-correction capabilities, demonstrating that building a system functioning at this level required careful adjustment to specific tasks, optimization of agent coordination, and fine-tuning of decision logic.
Governance controls
The attackers bypassed safety guardrails by framing the work as authorized penetration testing.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.