GitLab Fixes Critical AI Gateway Flaw Enabling Command Execution

Summary
GitLab fixed CVE-2026-90970, a critical AI Gateway flaw that could let an authenticated Duo Agent Platform user execute commands. Self-hosted gateway operators should update to versions 19.2.4, 19.3.2, or 19.4.1.
Key points
- The flaw affects GitLab’s self-hosted AI Gateway and is rated critical, with a CVSS score of 9.9.
- A logged-in user with Duo Agent Platform access could escape a custom flow’s prompt-template sandbox and execute arbitrary commands on the gateway.
- Fixed gateway versions are 19.2.4, 19.3.2, and 19.4.1; the article says no fixed version is listed below 19.2.4.
- GitLab recommends that organizations hosting their own gateway update immediately. GitLab-hosted gateway users do not need to act.
- The advisory does not describe whether the flaw has been exploited; CISA’s CVE assessment lists exploitation as “none.”
- No workaround is listed for gateways that cannot yet be updated.
Article Details
- Vulnerability Types
- Prompt-template sandbox escape leading to arbitrary command execution
- Template engine weakness (CWE-1336)
- Severity
- Critical; CVSS 9.9
- Affected Versions
- AI Gateway 18.1.6 through 19.2.3
- AI Gateway 19.3 before 19.3.2
- AI Gateway 19.4 before 19.4.1
- Exploitation Status
- not_reported
- Exploit Availability
- none_reported
- Patch Status
- available
CVE
Vendors
Products
AI GatewayGitLab has fixed a critical AI Gateway flaw, tracked as CVE-2026-90970, that could let a logged-in user with Duo Agent Platform access execute arbitrary commands under certain conditions.GitLab Duo Agent PlatformGitLab has fixed a critical AI Gateway flaw, tracked as CVE-2026-90970, that could let a logged-in user with Duo Agent Platform access execute arbitrary commands under certain conditions.
