Skip to main content
Vulnerability Database/CVE-2026-92470

CVE-2026-92470: GitLab Auth Bypass Vulnerability

CVE-2026-92470 is an authentication bypass flaw in GitLab Enterprise Edition that allows authenticated users to access sensitive CI/CD variables through Duo AI troubleshooting. This post covers technical details, affected versions, impact, and mitigation steps.

Updated:

CVE-2026-92470 Overview

CVE-2026-92470 is a missing authorization vulnerability [CWE-862] in GitLab Enterprise Edition (EE). The flaw lets an authenticated user retrieve sensitive CI/CD variable values from debug-mode job traces through the Duo AI troubleshooting feature. Affected releases include all versions from 18.7 before 19.2.7, 19.3 before 19.3.3, and 19.4 before 19.4.1. GitLab issued a patch in release 19.4.1 and corresponding backport versions. The vulnerability carries scope-changed confidentiality impact because exposed CI/CD variables often include credentials, API tokens, and signing keys used by downstream systems.

Critical Impact

Authenticated users can harvest secrets from debug job traces through the Duo AI troubleshooting feature, enabling lateral movement into cloud, registry, and deployment systems that trust those credentials.

Affected Products

  • GitLab EE versions 18.7 up to (excluding) 19.2.7
  • GitLab EE versions 19.3 up to (excluding) 19.3.3
  • GitLab EE versions 19.4 up to (excluding) 19.4.1

Discovery Timeline

  • 2026-09-24 - CVE-2026-92470 published to NVD
  • 2026-09-28 - Last updated in NVD database

Technical Details for CVE-2026-92470

Vulnerability Analysis

The vulnerability resides in the Duo AI troubleshooting feature, which assists users in diagnosing failed CI/CD jobs. When a job runs with CI_DEBUG_TRACE enabled, the job trace contains expanded environment variables, including masked and protected CI/CD variables. The Duo troubleshooting workflow reads these traces to generate AI-assisted explanations.

The affected code paths do not enforce the same authorization checks that normally gate direct access to debug job traces. As a result, an authenticated user who should not be able to view a protected job trace can still obtain the trace content indirectly through the AI troubleshooting response.

Because the exposed data includes production secrets, the impact extends beyond GitLab itself. Attackers can pivot into connected registries, cloud accounts, and deployment targets that trust the leaked credentials.

Root Cause

The root cause is a missing authorization check [CWE-862] on the server-side handler that supplies job trace content to the Duo AI troubleshooting feature. The feature trusts the requesting user without re-validating that they have permission to view the underlying job trace or its sensitive content.

Attack Vector

The attack requires network access to a GitLab EE instance and valid authenticated credentials with low privileges. The attacker triggers the Duo AI troubleshooting feature against a job that executed with debug tracing enabled. The feature returns trace content containing CI/CD variable values that the attacker would not otherwise be authorized to read. No user interaction is required by the victim, and the scope change reflects that leaked secrets authorize access to systems outside the GitLab trust boundary.

See the GitLab Patch Release Note and GitLab Work Item Overview for vendor technical details.

Detection Methods for CVE-2026-92470

Indicators of Compromise

  • Unexpected invocations of the Duo AI troubleshooting feature against jobs in projects where the calling user has no legitimate role.
  • Jobs executed with CI_DEBUG_TRACE set to true in projects that handle production secrets.
  • Reuse of CI/CD variable values (API tokens, cloud credentials) from external systems shortly after a Duo troubleshooting request.

Detection Strategies

  • Audit GitLab application logs for Duo AI troubleshooting requests correlated with the user, project, and target job ID.
  • Flag any troubleshooting access where the requesting user is not a developer, maintainer, or owner of the target project.
  • Review CI job configuration history for recent additions of CI_DEBUG_TRACE on pipelines that touch sensitive variables.

Monitoring Recommendations

  • Forward GitLab audit events and Rails production logs to a centralized analytics platform for cross-project correlation.
  • Monitor downstream systems (cloud IAM, container registries, package repositories) for anomalous authentication from leaked CI/CD credentials.
  • Alert on bulk or scripted calls to the Duo troubleshooting endpoint from a single session.

How to Mitigate CVE-2026-92470

Immediate Actions Required

  • Upgrade GitLab EE to 19.4.1, 19.3.3, or 19.2.7 depending on the deployed branch.
  • Rotate every CI/CD variable that may have appeared in a debug-mode job trace, including deployment tokens, cloud credentials, and signing keys.
  • Review Duo AI troubleshooting usage across the instance and investigate any requests that cannot be tied to a legitimate project member.

Patch Information

GitLab released fixed versions 19.4.1, 19.3.3, and 19.2.7. Full details are available in the GitLab Patch Release Note. Self-managed administrators should apply the patch promptly; GitLab.com has been updated by the vendor.

Workarounds

  • Disable the GitLab Duo AI troubleshooting feature at the instance or group level until the patch is applied.
  • Remove CI_DEBUG_TRACE from pipeline configurations and project or group variables to prevent expansion of secrets into job traces.
  • Restrict project membership and job trace visibility settings so that only required users can access CI/CD metadata.
bash
# Configuration example: ensure debug tracing is disabled project-wide
# In .gitlab-ci.yml, remove or set to false:
variables:
  CI_DEBUG_TRACE: "false"

# Restrict job log visibility in project settings
# Settings > CI/CD > General pipelines > Public pipelines: disabled

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Experience the Most Advanced Cybersecurity Platform

See how the world’s most intelligent, autonomous cybersecurity platform can protect your organization today and into the future.