CVE-2026-83495

OpenShift AI Trainer jobs could reveal namespace secrets beyond user permissions

Red Hat · Red Hat OpenShift AI Trainer v2

First documented disclosure · Review checkpoint

Summary

A job author could arrange for the workload to consume a namespace secret outside the author’s read permissions.

Severity and exploitation evidence

Low · numeric CVSS score not recorded. Severity source.

Exploitation not established: Malicious exploitation status remains unverified in the reviewed sources.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

Environment-variable secret references inherited broader workload permissions.

Conditions required

TrainJob creation permissions exceed the user’s permission to read namespace secrets.

Potential impact

Sensitive values in the same namespace can be disclosed through the training workload.

Affected versions and configurations

Trainer v2 Technology Preview in Red Hat OpenShift AI. Version applicability remains under investigation.

Fix and mitigation status

Red Hat reports the affected package as under investigation.

Practical next steps

Align TrainJob creation with Secret-read permissions or restrict permitted secret names in a validating webhook.

Operating lesson

Constrain the privileges a job can exercise on behalf of its author.

Evidence and limits

Original disclosure 2026-08-31. The CVE Program published this record 2026-11-07. Red Hat supplies an August 31 public date and Low severity, citing Technology Preview status and the required RBAC mismatch.

Sources