Skip to main content
UNIFIED SOC INCIDENT RESPONSE BLUEPRINT

GitLab AI Gateway Critical RCE Flaw Allows Remote Code Execution

GitLab patches a critical RCE flaw in its AI Gateway service enabling authenticated users to escape prompt sandboxes and run arbitrary code on hosting servers.

Detection & Hunting Detectors (1)

[T1190] GitLab AI Gateway Critical RCE Flaw Allows Remote Code Execution Hunt QueryStandalone Rule
// Hunt for abnormal child process creation from GitLab AI Gateway container workloads
DeviceProcessEvents
| where Timestamp > ago(7d)
| where ProcessCommandLine has_any ("ai_gateway", "duo_agent", "gitlab-ai")
    or FolderPath has_any ("/opt/gitlab", "/ai-gateway")
| where FileName in~ ("bash", "sh", "dash", "curl", "wget", "python", "python3", "nc")
| where InitiatingProcessFileName in~ ("python", "python3", "uvicorn", "gunicorn")
| project Timestamp, DeviceName, InitiatingProcessFileName, InitiatingProcessCommandLine, FileName, ProcessCommandLine, AccountName
| order by Timestamp desc

Verified Indicators of Compromise (2)

Indicators of Compromise (IOCs)
2 Identified
COMPONENTDuo Agent Platform Prompt Flow Engine
SERVICEGitLab AI Gateway (Standalone Service)