Chinese hackers used AI agent Hermes in Thai Ministry of Finance cyberattack
Chinese attackers used the AI agent Hermes v3 operating in autonomous mode to breach Thailand's Ministry of Finance, gaining access to 47 computers, 54.9 TB of financial data, SQL databases, and biometric systems. Security researchers Hunt.io and Bob Diachenko discovered and reported the attack on July 15, 2026.
What happened
Chinese hackers used the AI agent Hermes v3 to penetrate Thailand's Ministry of Finance through a PHP-based Webshell on an internet-facing web server.
The AI agent operated in autonomous YOLO mode to identify targets, move laterally across the network, and exploit known vulnerabilities including CVE-2021-4034 (PwnKit), CVE-2021-3156 (sudo vulnerability), CVE-2017-7269 (IIS vulnerability), and newer exploits such as CVE-2026-31431 (Copy Fail), CVE-2026-43284, CVE-2026-43500 (Dirty Frag), and CVE-2026-43503 (DirtyClone).
Hunt.io and researcher Bob Diachenko discovered event logs left by Hermes showing the AI agent identifying Ministry hosts, traversing files, and extracting LinPEAS output. The attackers deployed a Go-language malware called Hades within the Ministry's network. Hunt.io and Diachenko reported the incident to Thailand CERT and NCSA on July 15, 2026.
Reported impact
The attack provided hackers with access to at least 47 Ministry of Finance computers and servers, 54.9 TB of government financial data, 14 Microsoft SQL databases, biometric systems, and Remote Desktop Gateway infrastructure.
Index analysis
The attack represents one of the first confirmed uses of autonomous AI agents in cyberattacks against a government entity. The AI agent's YOLO mode allowed it to operate without human intervention in identifying and exploiting vulnerabilities.
The incident reflects broader trends in AI-accelerated cyberattacks, including faster exploitation timelines and the emergence of freely available models capable of advanced network reconnaissance, as demonstrated by earlier AI agent research.
Governance controls
The researchers discovered the attack infrastructure through three public folders on a server in Hong Kong (43.246.208[.]207) containing CVE exploit code, Webshells, Suo5 HTTP tunnels, and scripts targeting email and Apache Hadoop infrastructure, totalling 585 files of approximately 470 MB.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.