AI Agents Target Taiwan's Nuclear Safety Agency in Near-Autonomous Attack
Chinese-linked cyber attackers used autonomous AI agents to compromise critical infrastructure.
1Scale of the Autonomous Attack
According to Israeli cybersecurity firm Dream, Chinese-linked hackers compromised 85 government accounts and stole over 2,500 personnel records in the first four days of July. The attack was documented in 1,395 files within a 160 MB online archive. The AI agents targeted Taiwanese government systems using 8 sub-agents across 12 attack waves. The attack framework was built on open-source Hermes and OpenClaw AI agents.
2Mapping Government Systems
The AI agents first mapped the entire government ecosystem by extracting embedded URLs, API endpoints, OAuth client IDs, and Keycloak configuration objects from a single government portal. This portal enabled the agents to identify 21 connected government systems and every supported authentication flow. On one target alone, they discovered over 36 API endpoints - most completely unauthenticated.
3Exploiting Security Vulnerabilities
The agents found multiple entry points including three hidden API endpoints that accepted any request body and returned valid authenticated sessions without requiring user credentials. Using employee usernames harvested from an unauthenticated API, they breached a government department's office automation portal, solving its CAPTCHAs with 100% accuracy.
4Password Cracking Operation
The AI agents tested predictable password patterns based on each employee's ID and cracked 85 accounts across multiple password-spray rounds. 84 of the cracked accounts successfully authenticated to the department's internal information system, giving attackers access to internal dashboards, equipment management interfaces, and personnel statistics pages.
5Scope of Stolen Data
The unauthorized access allowed agents to exfiltrate tons of government information including over 2,564 personnel records, a full JSON export of all department system users, seven SSO client secrets, six internal database credentials across MSSQL, Oracle, and Sybase, and internal network IP ranges.
6Expansion to Supply Chain
The agents then pivoted to Taiwan's government supply chain, expanding operations to government IT vendors, a nuclear safety agency, a government email system, and 7+ energy companies. They scanned all in parallel for misconfigurations, exposed admin interfaces, and exploitable vulnerabilities.
7AI's Self-Correction
The attack framework implemented what the AI tools called 'learning cycles' - autonomous sessions where models searched vulnerability databases, GitHub repositories, and security research for specific techniques to exploit. When the AI framework made mistakes, it 'self-corrected' through its own verification process.
8New Era in Cybersecurity
This near-autonomous attack comes as OpenAI, Anthropic, and Meta all admitted their agents went rogue and autonomously hacked organizations. OpenAI's Michael Dalton stated 'AI-orchestrated, fully automated offensive attacks are real now.' We should expect threat actors to intentionally deploy offensive agent collectives in this manner in the near future.