RELIAQUEST Threat Spotlight has documented an AI-driven attack that turned a single exposed endpoint into a full server takeover in under 24 hours. The intrusion entered through an Apache Tomcat server running a Spring Batch application that accepted job requests with no login. The attacker had pre-staged Java exploit material and a commercial vulnerability scanner on the host, enabling a rapid escalation once the exposed job submission feature was hit. Crucially, no new malware or zero-day was involved; the operation relied on existing tools and in‑app code execution rather than external payloads.
The infection chain shows the attacker submitting jobs that invoked Nashorn, the JavaScript engine built into Java, which executed within the application and did not spawn new processes, delaying traditional process-based detection. The attacker then extracted plaintext database credentials from configuration files with SQL Server admin rights, enabled xp_cmdshell to run OS commands, and leveraged SeImpersonatePrivilege to gain SYSTEM access.
PrintSpoofer and GodPotato tools were staged in advance, with a final step creating local admin accounts and harvesting registry hives (SAM, SYSTEM, SECURITY) for offline password cracking. Command and control occurred through the job endpoint, while a separate address trafficked results in bulk. The ReliaQuest team notes the AI agents performed many steps with minimal human input, indicating a speed‑driven breach that defenders must detect by behaviour and automation patterns rather than single-tool signatures.
Defensive recommendations include enforcing authentication on exposed endpoints, retaining job histories, disallowing plaintext credentials, disabling xp_cmdshell, and alerting on privilege escalations and registry exports.