CVE-2026-18942: Feast Operator Privilege Escalation Flaw

CVE-2026-18942 Overview

CVE-2026-18942 is a code injection vulnerability [CWE-94] in the Feast operator, a Kubernetes operator that manages Feast feature store deployments. A malicious tenant can inject arbitrary code into their feature repository. An automated process running with elevated privileges then executes that code, exposing sensitive credentials to the attacker. The attacker can leverage those stolen credentials to escalate privileges and gain administrative control over the entire Kubernetes cluster. Red Hat published advisories RHSA-2026:53261 and RHSA-2026:53262 addressing the flaw.

Critical Impact
A tenant with permission to modify a feature repository can pivot from tenant-level access to full administrative control of the Kubernetes cluster through credential theft.

Affected Products

Feast operator (Feature Store for machine learning) as distributed by Red Hat
Deployments referenced in Red Hat Security Advisory RHSA-2026:53261
Deployments referenced in Red Hat Security Advisory RHSA-2026:53262

Discovery Timeline

2026-08-10 - CVE-2026-18942 published to the National Vulnerability Database
2026-08-12 - Last updated in NVD database

Technical Details for CVE-2026-18942

Vulnerability Analysis

The Feast operator automates reconciliation of feature repository content submitted by tenants. During reconciliation, the operator processes tenant-controlled files inside a privileged automation context. Because tenant input is treated as trusted logic rather than data, a malicious tenant can embed executable code in their feature repository. The operator's automated pipeline evaluates that code with elevated service account privileges. Once executing, the injected code can read Kubernetes secrets, service account tokens, and other credentials mounted into the operator's execution environment. Those credentials typically grant broad cluster-scoped rights, so a successful injection escalates from tenant scope to cluster administrator. The Common Weakness Enumeration classification is [CWE-94] Improper Control of Generation of Code (Code Injection).

Root Cause

The operator fails to enforce a trust boundary between tenant-supplied feature repository content and its own execution context. Tenant input is not sandboxed, validated as data-only, or executed under a least-privileged service account.

Attack Vector

An authenticated tenant with write access to their feature repository submits a repository containing malicious code. The Feast operator reconciles the repository and executes the payload with its own service account privileges. The payload extracts secrets from the pod environment and uses them to authenticate to the Kubernetes API as a cluster administrator.

// No verified exploit code is publicly available.
// Refer to the Red Hat advisories for technical remediation details.

Detection Methods for CVE-2026-18942

Indicators of Compromise

Unexpected outbound network connections originating from the Feast operator pod to external endpoints not tied to feature store data sources.
Kubernetes API audit log entries showing the Feast operator service account performing actions outside routine reconciliation, such as reading secrets in unrelated namespaces.
New or modified ClusterRoleBinding or RoleBinding objects created shortly after a tenant push to a feature repository.
Feature repository commits containing executable code, shell invocations, or references to Kubernetes secret paths such as /var/run/secrets/kubernetes.io/serviceaccount/token.

Detection Strategies

Enable Kubernetes API server audit logging and alert on the Feast operator service account accessing secrets, serviceaccounts/token, or RBAC resources.
Monitor container runtime telemetry for the Feast operator pod, flagging process executions outside the expected operator binary set.
Review feature repository commit history for code additions inside files that should hold declarative feature definitions only.

Monitoring Recommendations

Baseline the Feast operator's normal API calls and network destinations, then alert on statistical deviations.
Forward Kubernetes audit logs and container runtime events to a centralized analytics platform for correlation across tenants.
Track service account token usage to detect reuse of the operator's token from unexpected client IP ranges or user agents.

How to Mitigate CVE-2026-18942

Immediate Actions Required

Apply the fixes referenced in Red Hat Security Advisory RHSA-2026:53261 and Red Hat Security Advisory RHSA-2026:53262.
Rotate all Kubernetes secrets, service account tokens, and cloud credentials that were reachable from the Feast operator pod.
Audit RBAC bindings created since the operator was deployed and remove any that were not authorized.
Restrict which tenants and identities can push to feature repositories consumed by the operator.

Patch Information

Red Hat released fixes tracked in RHSA-2026:53261, RHSA-2026:53262, and documented in the Red Hat CVE page for CVE-2026-18942. See Red Hat Bug Report #2511118 for engineering context. Upgrade the Feast operator to the fixed version distributed through these advisories.

Workarounds

Reduce the Feast operator service account to the minimum RBAC needed for reconciliation and remove cluster-admin equivalents.
Isolate the operator in a dedicated namespace with NetworkPolicy rules blocking egress to non-required destinations.
Require code review and signed commits on feature repositories before the operator ingests them.

bash

# Example: restrict egress from the Feast operator namespace
kubectl apply -f - <<EOF
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: feast-operator-egress
  namespace: feast-system
spec:
  podSelector:
    matchLabels:
      app: feast-operator
  policyTypes:
  - Egress
  egress:
  - to:
    - namespaceSelector:
        matchLabels:
          name: kube-system
    ports:
    - protocol: TCP
      port: 443
EOF