GitLab Patches Critical AI Gateway Vulnerability Enabling Command Execution

· Original article ↗

Summary

CVE-2026-90970 lets an authenticated user with Duo Agent Platform access execute commands on affected Self-Hosted AI Gateways. GitLab released fixes; exploitation is unconfirmed.

Key points

  • The template-engine flaw can let an authenticated user with Duo Agent Platform access escape the prompt-template sandbox and execute arbitrary commands on the gateway.
  • GitLab rates the vulnerability CVSS 3.1 9.9; active exploitation has not been confirmed.
  • Affected releases are fixed in 19.2.4, 19.3.2, or 19.4.1, as applicable. Verify the gateway image or Helm deployment, which may be separate from the GitLab application.
  • GitLab-hosted AI Gateways have already been patched; customers using them do not need to take action for this issue.
  • Restrict access to Duo Agent Platform and custom flows, and limit gateway network access; these measures do not replace upgrading.
  • Investigate unexpected flow changes, child processes, access to mounted files, and outbound connections. Rotate potentially exposed credentials if suspicious command execution is found.

Article Details

Vulnerability Types
  • Template engine vulnerability
Severity
Critical; CVSS 3.1 score 9.9
Affected Versions
  • Self-Hosted AI Gateway 18.1.6 through before 19.2.4
  • Self-Hosted AI Gateway 19.3 through before 19.3.2
  • Self-Hosted AI Gateway 19.4 through before 19.4.1
Exploitation Status
not_reported
Exploit Availability
none_reported
Patch Status
available
Workarounds
  • Limit Duo Agent Platform and custom flow access to users who require it.
  • Restrict gateway network access to required services and trusted administrative networks; these measures do not replace the update.

CVE

Vendors

Products

Related Articles