Why ArgoCD 2.13 security risks matter for FluxCD and GitLab users
ArgoCD faces critical security vulnerabilities like CVE-2025-55190 and credential exposure risks in versions 2.13.0 through 2.13.8. These flaws contrast with FluxCD's decentralized architecture, which avoids the massive blast radius inherent in ArgoCD's hub-and-spoke model.
ArgoCD uses a hub-and-spoke model. A centralized control plane in a management cluster governs multiple spoke clusters across different regions. This architecture provides a single pane of glass for visibility and governance. Platform teams enforce global policies and manage multi-cluster RBAC from one dashboard. However, the management cluster needs high-level credentials for every production cluster it manages. This concentration of credentials creates a massive blast radius in the current threat landscape. ArgoCD can manage 50 or more clusters from a single instance. FluxCD uses a decentralized toolkit of independent controllers in each target cluster. Each cluster pulls its own configuration from Git or OCI repositories without needing an external coordinator. This approach avoids the central bottleneck and cross-cluster credential risk inherent in the hub-spoke model. FluxCD is the preferred choice for edge computing because it operates with minimal resource overhead and no inbound network requirements.
| Specification | FluxCD | ArgoCD |
|---|---|---|
| Control Plane | Per-cluster agents | Hub-and-Spoke |
| Idle Memory | ~200 MB | ~500 MB |
| 100 Apps Memory | +50-100 MB | +200-400 MB |
| UI | CLI-first | Web Dashboard |
ArgoCD 3.3 updates and the security reality
ArgoCD 3.3 addresses operational gaps for mature organizations. It introduces PreDelete hooks to manage application lifecycles. ArgoCD 3.3 introduces PreDelete hooks to allow teams to define Kubernetes resources, such as specialized Jobs, that must execute and succeed before ArgoCD removes the rest of an application’s manifests from the cluster. This capability allows teams to perform data exports or traffic draining in service meshes. The release also includes OIDC background token refresh to stop session timeouts. Because shallow cloning support reduces Git fetch times in large monorepos from minutes to seconds, the efficiency of large-scale syncs increases significantly. The Source Hydrator also uses Git notes to track hydration state instead of creating new commits. This decreases repository bloat. ArgoCD can support up to 50,000 applications with proper tuning. The ApplicationSet controller also allows for progressive rollouts using a RollingSync strategy.
Security remains a major problem for ArgoCD users. API tokens with project-level permissions in ArgoCD versions 2.13.0 through 2.13.8 can retrieve sensitive repository credentials. This flaw allows attackers to access usernames and passwords for private repositories. I find this security gap unacceptable. ArgoCD also faces severe Denial of Service risks. Malicious API requests can crash the API server if a webhook.bitbucketserver.secret is not configured. An attacker can also exploit unsafe array manipulation in a multi-threaded environment to crash the application. The critical CVE-2025-55190 vulnerability allows attackers to bypass brute force login protection. The high-severity CVE-2024-21661 flaw allows an attacker to initiate a Denial of Service attack via unsafe array manipulation. ArgoCD’s popularity is high. It has roughly 23,100 GitHub stars as of August 2026.
Flux 2.8 capabilities and GitLab integration
Flux 2.8 changes the visibility landscape with the Flux Operator Web UI. This dashboard provides workload monitoring and synchronization statistics. It supports OIDC and Kubernetes RBAC out of the box. Flux 2.8 also uses Helm v4 with Server-Side Apply to improve drift detection. This reduces the conflict storms seen when multiple controllers manage the same resource. Flux also uses kstatus-based health checking for HelmRelease objects. It supports CEL-based health check expressions for complex readiness logic. Flux 2.8 also introduces interruptible reconciliation to reduce Mean Time to Recovery. This mechanism allows operators to cancel ongoing health checks and trigger a new reconciliation as soon as a fix appears in Git. Flux’s image automation detects new image tags and creates Git commits to update manifests.
ArgoCD and FluxCD handle Helm differently. ArgoCD performs a helm template on its repository server to create plain Kubernetes YAML manifests. This provides manifest transparency. FluxCD uses the Helm SDK to perform native helm install and helm upgrade operations. This means Flux-managed applications appear in standard helm list commands.
GitLab CI/CD works with both tools. Teams use GitLab CI/CD to build images and push them to a registry. GitLab CI/CD then triggers the GitOps controller to sync the cluster. You can use Kubernetes runners to run your pipelines. GitLab CI/CD uses .gitlab-ci.yml to define pipelines. Flux has roughly 8,180 GitHub stars as of July 2026. The project has reached 210 releases and more than 30.2 billion cumulative image downloads.
Does the Flux Web UI provide enough depth to replace the ArgoCD dashboard for complex enterprise workflows? You should skip ArgoCD if your team requires a lightweight, CLI-only footprint.