Between July 9 and July 13, 2026, threat intelligence researchers discovered exposed directories on attacker-controlled servers containing extensive artifacts related to an operation targeting Thailand's Ministry of Finance. These included exploit code, web shells, stolen credentials, and logs from the Hermes AI agent, an open-source autonomous AI tool running in YOLO mode that executes commands without human intervention. The Hermes agent was used to automate post-exploitation tasks such as privilege escalation, kernel vulnerability scanning, service enumeration, and file system searches. The attackers targeted internal services including Hadoop, Apache Ambari, GlassFish, and mail servers. While the Ministry of Finance has not confirmed a breach, recovered artifacts indicate active intrusion and lateral movement. The initial access vector remains unknown. This case exemplifies the increasing use of AI agents to automate complex cyberattack operations.

The attack resulted in unauthorized access to multiple internal systems within the Ministry of Finance's network, deployment of web shells, and execution of automated post-exploitation activities. The attackers gathered sensitive information such as personnel records and performance assessments but there is no confirmed evidence of data exfiltration. The use of an autonomous AI agent in YOLO mode allowed the threat actor to conduct extensive reconnaissance and privilege escalation without continuous human oversight, increasing the efficiency and scale of the intrusion. The incident demonstrates a medium-level impact due to partial compromise and potential exposure of sensitive internal data, though full breach confirmation and data loss have not been established.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Organizations should verify and secure exposed infrastructure to prevent unauthorized access. Monitoring for web shells and unusual automated activity is recommended. Since the Hermes AI agent operated in unattended mode, restricting execution of unauthorized automation tools and enforcing strict access controls can reduce risk. Coordination with national cybersecurity agencies, such as ThaiCERT, is advised for incident response and threat intelligence sharing. No official fix or patch is applicable to the AI agent itself as it is an open-source tool used by attackers; mitigation focuses on detection and prevention of unauthorized access and lateral movement.