---
id: "CVE-2026-83495"
page: "https://infrastack.ai/live/CVE-2026-83495"
kind: "vulnerability"
aliases: ["CVE-2026-83495"]
title: "Kubeflow Trainer: TrainJob users can read namespace secrets via env injection"
severity: "low"
cvss: null
epss: null
exploited: false
status: "open"
category: "infrastructure"
products: [{"name": "Red Hat OpenShift AI (Trainer v2, tech preview)", "fixed": [], "affected": "unknown", "ecosystem": "vendor"}]
ecosystems: []
tags: ["infrastructure", "kubeflow", "kubernetes", "openshift-ai", "rbac", "self-hosted", "trainer-v2"]
not_affected_if: ["no-trainer-v2", "no-trainjob-creators-untrusted", "secret-rbac-matches-trainjob-rbac", "trainer-not-installed", "trainjob-creators-can-read-secrets"]
published_at: "2026-11-07T07:17:04+00:00"
first_seen_at: "2026-11-07T07:46:07+00:00"
updated_at: "2026-11-07T07:54:21+00:00"
sources: [{"source": "nvd", "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-83495", "first_seen_at": "2026-11-07T07:46:07+00:00"}, {"source": "analyst", "url": "https://access.redhat.com/security/cve/CVE-2026-83495", "first_seen_at": "2026-11-07T07:48:47+00:00", "cites": ["https://access.redhat.com/security/cve/CVE-2026-83495", "https://app.opencve.io/cve/CVE-2026-83495", "https://access.redhat.com/security/cve/CVE-2026-83495"]}]
exploit_sources: []
---

# Kubeflow Trainer: TrainJob users can read namespace secrets via env injection

Link for people: https://infrastack.ai/live/CVE-2026-83495 (this .md is for agents)

Trainer v2 accepts valueFrom.secretKeyRef in TrainJob env definitions, so a TrainJob creator can read same-namespace Secrets beyond their RBAC. Red Hat mitigation: align TrainJob RBAC with Secret read RBAC, or restrict secretKeyRef in the validating webhook to an allowlist of secret names. No fixed version published yet.

_Summary, check and fix written by a model from the sources; labelled as such._

## Affected

- vendor:Red Hat OpenShift AI (Trainer v2, tech preview) affected unknown fixed no fixed version listed

## How to check

Check whether Kubeflow Trainer v2 is installed (kubectl get crd trainjobs.trainer.kubeflow.org; on OpenShift AI, whether the Trainer v2 tech preview is enabled). Then compare who can create TrainJobs (kubectl auth can-i create trainjobs --as <user> -n <ns>) with who can read Secrets there; users who can create TrainJobs but not read Secrets are the exposure. Affected versions are not published yet.

## Timeline

- 2026-11-07T07:46:07+00:00 nvd https://nvd.nist.gov/vuln/detail/CVE-2026-83495
- 2026-11-07T07:48:47+00:00 analyst https://access.redhat.com/security/cve/CVE-2026-83495 — Read Red Hat CVE page (vendor, assigner). Red Hat rates Low; Trainer v2 is a Tech Preview in OpenShift AI. No affected/fixed versions published anywhere yet; upstream kubeflow/trainer advisory not found. Severity lowered to Red Hat's own Low rating.

## Documents

- [advisory] Red Hat CVE-2026-83495 https://access.redhat.com/security/cve/CVE-2026-83495
- [mentions] OpenCVE CVE-2026-83495 https://app.opencve.io/cve/CVE-2026-83495
- [mitigation] Red Hat CVE-2026-83495 https://access.redhat.com/security/cve/CVE-2026-83495

## From the sources

A flaw was found in the Trainer v2 TrainJob API. A user with permissions to create TrainJobs can exploit this by injecting `valueFrom.secretKeyRef` into environment variable definitions. This allows them to read and exfiltrate sensitive information (secrets) from the namespace, leading to a privilege escalation. This occurs when the TrainJob Role-Based Access Control (RBAC) is broader than the Secret read RBAC.

---

_Free to read, with fair use of about 60 pages a minute. For higher limits or API access: sales@infrastack.ai_