CVE-2026-18942 is a vulnerability in the Feast operator component of Red Hat OpenShift AI (RHOAI). A malicious tenant can inject arbitrary Python code into their feature repository, which is then executed by the feast apply CronJob using the feature-server's ServiceAccount permissions. This execution context allows the attacker to access the ServiceAccount token and escalate privileges to cluster-admin level, potentially gaining full administrative control over the Kubernetes cluster. The vulnerability is due to the feast apply CronJob running untrusted code with elevated privileges. Red Hat advises configuring the Feast operator to use a dedicated, minimally-privileged ServiceAccount with automountServiceAccountToken disabled, applying restricted security contexts to the CronJob pod, and limiting network egress to required registry endpoints.

The vulnerability allows a malicious tenant to execute arbitrary code with elevated privileges within the Kubernetes cluster, leading to theft of sensitive credentials and escalation to cluster-admin privileges. This can result in full administrative control over the Kubernetes environment, posing a significant security risk to the cluster and its workloads.

Red Hat recommends configuring the Feast operator to use a dedicated, minimally-privileged ServiceAccount for the feast apply CronJob, ensuring that automountServiceAccountToken is set to false. Additionally, apply a restricted securityContext to the CronJob pod and implement network policies to restrict egress traffic to only necessary registry endpoints. Ensure that the CronJob ServiceAccount is separate from the feature-server ServiceAccount. These mitigations reduce the risk of privilege escalation by limiting the permissions and exposure of the automated process executing tenant code.