Improper Neutralization of Special Elements in Output Used by a Downstream Component in GitLab AI Gateway - CVE-2026-90970
Published: October 5, 2026
Vulnerability details
The vulnerability allows a remote user to execute arbitrary commands on the AI Gateway.
The vulnerability exists due to improper neutralization in the custom flow prompt template when processing a specially crafted flow configuration. A remote user can escape the prompt template sandbox to execute arbitrary commands on the AI Gateway.
Exploitation requires Duo Agent Platform access.