How AI Agents Turned One Exposed Endpoint into a Server Takeover
An investigation revealed the first confirmed incident where large language model-driven AI agents executed substantial portions of an attack. Within 24 hours, attackers exploited an exposed job-submission feature on an Apache Tomcat server running Spring Batch, requiring no authentication, zero-days, or new malware. The attacker used Cairn, an open-source AI agent orchestration platform, identified on the attacking infrastructure. Commands were executed through JavaScript via Nashorn engine, with results extracted through error messages in 1,800-byte chunks across hundreds of sequential requests. The attacker recovered plaintext database credentials from configuration files, enabled xp_cmdshell, and escalated privileges using PrintSpoofer and GodPotato tools to achieve SYSTEM access. Registry hives were extracted, administrator accounts created, and artifacts deleted. Machine-driven execution was evidenced by rapid command timing, programmatic job naming, structured output formatting, and automated error ...
Pulse ID: 6ad07b924dde0538386bd7d5
Pulse Link: https://otx.alienvault.com/pulse/6ad07b924dde0538386bd7d5
Pulse Author: AlienVault
Created: 2026-10-15 07:06:58
Be advised, this data is unverified and should be considered preliminary. Always do further verification.
