// Security Advisory Triage
GitLab AI Gateway CVE-2026-90970: arbitrary command execution via prompt template escape
GitLab patch releases · published 2026-10-03 01:36 UTC
WHAT CHANGEDGitLab released patches 19.2.4, 19.3.2, and 19.4.1 for AI Gateway on 2026-10-02. CVE-2026-90970 (CVSS 9.9 critical) allows authenticated users with Duo Agent Platform access to escape prompt template sandbox via crafted flow configuration, achieving arbitrary command execution. Affects all versions from 18.1.6 before 19.2.4, 19.3 before 19.3.2, and 19.4 before 19.4.1. Self-hosted installations only; GitLab.com, Dedicated, and instances using GitLab-hosted gateway already patched.
WHY IT MATTERSAn authenticated attacker with low-privilege Duo Agent access can execute arbitrary commands on your AI Gateway server. CVSS 9.9 reflects network-accessible, low-complexity exploitation with complete system compromise (confidentiality, integrity, availability all rated High, scope Changed). Self-hosted gateways remain exposed until patched.
WHAT TO DOIdentify all self-hosted AI Gateway instances running 18.1.6 through 19.4.0. Upgrade to 19.2.4, 19.3.2, or 19.4.1 immediately following the install documentation at docs.gitlab.com. If you use GitLab.com, Dedicated, or a GitLab-hosted gateway, no action required — already patched.
Model-reported confidence 95% (not measured accuracy)
Primary source ↗
written by anthropic/claude-sonnet-4.5
Show the reasoning — raw diff, triage verdict, confidence
RAW DIFF
+## GitLab AI Gateway Critical Patch Release: 19.2.4, 19.3.2, and 19.4.1 +https://docs.gitlab.com/releases/patches/other-patches/patch-release-gitlab-ai-gateway-19-4-1-released/ +2026-10-02T00:00:00Z + +We have released versions 19.2.4, 19.3.2, and 19.4.1 of the GitLab AI Gateway. +These versions contain a critical security fix for GitLab Self-Hosted AI Gateway, and we strongly recommend that all GitLab Self-Managed customers with GitLab Self-Hosted AI Gateway installations update to one of these versions immediately. We have conducted targeted outreach to Self-Hosted AI Gateway customers prior to this release post with this guidance. +A fix has already been deployed for GitLab-hosted AI Gateways. Customers using GitLab.com, GitLab Dedicated, and GitLab Self-Managed instances using a GitLab-hosted AI Gateway are protected and do not need to take action. +Recommended Action +We strongly recommend that all GitLab Self-Hosted AI Gateway installations running a version affected by the issue described below are upgraded to one of the patched versions listed above as soon as possible. +Security fixes +Table of security fixes + +Title +Severity + +Improper Neutralization issue in custom flow prompt template impacts AI Gateway +Critical + +CVE-2026-90970 - Improper Neutralization issue in custom flow prompt template impacts AI Gateway +GitLab has remediated an issue in the GitLab AI Gateway that, under certain conditions, could have allowed an authenticated user with Duo Agent Platform access to escape the prompt template sandbox via a specially crafted flow configuration, leading to arbitrary command execution on the AI Gateway. +Impacted Versions: GitLab AI Gateway: all versions from 18.1.6 before 19.2.4, 19.3 before 19.3.2, and 19.4 before 19.4.1 +CVSS 9.9 (CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H) +Thanks invisiblemeerkat for responsibly disclosing this issue. +Updating +To update GitLab Self-Hosted AI Gateway, see the GitLab Self-Hosted AI Gateway install documentation. +Receive Patch Notifications +To receive release notifications via RSS, subscribe to our patch release RSS feed or our RSS feed for all releases.
HOW IT WAS JUDGED
Triage signals below are model-reported assertions, preserved verbatim; they are not independent arithmetic verification.
| triage verdict | MATERIAL at 0.98, needed 0.5 |
|---|---|
| triage reasoning | Critical AI Gateway vulnerability enabling arbitrary command execution on self-hosted deployments requires immediate patching. |
| signals | critical severity rating · known vulnerability with CVE assigned · arbitrary command execution capability · AI provider SDK/gateway affected · authenticated but low barrier to exploitation · immediate patching recommended · self-hosted deployments require action |
| suppression | — |
| triage model | anthropic/claude-haiku-4.5 ($0.00000) |
| writer model | anthropic/claude-sonnet-4.5 ($0.00000) |