Executive brief
Argo Workflows is an automation engine used to run complex parallel jobs on Kubernetes clusters. A vulnerability allows any user with permission to create workflows to crash the central controller by providing a malformed configuration. This results in a persistent denial-of-service where all workflow processing across the entire cluster is halted until the malicious entry is manually removed.
Technical details
An improper validation of an array index (CWE-129) exists in the Argo Workflows controller's pod informer. Specifically, the podGCFromPod() function splits the 'workflows.argoproj.io/pod-gc-strategy' annotation value by a forward slash and unconditionally attempts to access the second element of the resulting array. If a user submits a workflow with an annotation lacking a slash, the controller triggers an out-of-bounds panic. Because this panic occurs within an informer goroutine outside the main recovery scope, the entire controller process crashes. The malicious pod remains in the cluster, leading to a persistent CrashLoopBackOff state that prevents any other workflows from being processed. The issue is resolved in versions 3.7.14 and 4.0.5.
Affected products
- argoproj Argo Workflows >= 3.6.5, <= 3.6.19; >= 3.7.0, <= 3.7.13; >= 4.0.0, <= 4.0.4
- Red Hat Red Hat OpenShift AI (RHOAI)
Timeline
- 2026-04-23: disclosed
- 2026-04-23: advisory
- 2026-04-23: patched