Skip to main content
Start freeLog in
← Back to blog
Developer Security1 min read

GitLab Urges Immediate Patch for Critical AI Gateway RCE (CVE-2026-90970)

GitLab disclosed a critical flaw in its AI Gateway that can allow command execution in self-hosted deployments under specific conditions. Security teams running self-managed gateways should prioritize upgrades, validate exposure, and monitor authentication and gateway logs for suspicious activity.

#gitlab#cve#aisecurity#devsecops#vulnerability#patchmanagement#appsec
GitLab Urges Immediate Patch for Critical AI Gateway RCE (CVE-2026-90970)

GitLab AI Gateway Critical RCE: What Self-Managed Teams Need to Do Now

What happened

GitLab warned of a critical vulnerability in AI Gateway tracked as CVE-2026-90970. Under certain conditions, a logged-in user with Duo Agent Platform access could execute commands on a vulnerable self-hosted gateway.

Who is affected

Organizations that run self-managed GitLab environments and host their own AI Gateway are in scope. GitLab-hosted gateway users were reported as already remediated by GitLab.

Why it matters

AI integration points are now part of the production attack surface. A compromise at the gateway layer can impact confidentiality of prompts and outputs, service integrity, and potentially downstream development workflows.

Technical details

  • CVE: CVE-2026-90970
  • Severity: Critical (publicly reported as CVSS 9.9)
  • Exposure: Self-hosted AI Gateway deployments
  • Exploitation status: No broad exploitation campaign publicly confirmed at publication time
  • Fixed versions: Public reports indicate patched AI Gateway releases were issued on October 2, 2026

Recommended actions

  • Upgrade AI Gateway to vendor-fixed releases immediately.
  • Inventory all GitLab AI integrations and confirm whether any environment is self-hosting the gateway.
  • Restrict access to Duo Agent Platform features to least privilege.
  • Review gateway, auth, and API logs for unusual command execution or anomalous request patterns.
  • Add temporary compensating controls such as network segmentation and stricter allowlists around gateway hosts until patching is complete.

Sources

  • https://www.bleepingcomputer.com/news/security/gitlab-warns-of-critical-rce-vulnerability-in-ai-gateway-service/
  • https://thehackernews.com/2026/10/gitlab-patches-critical-self-hosted-ai.html