GitLab Urges Immediate Patching of Critical AI Gateway RCE Flaw

GitLab warns customers to immediately patch a critical vulnerability, tracked as CVE-2026-90970, in its AI Gateway service that could allow attackers to run arbitrary commands on vulnerable instances. The flaw stems from an improper neutralization weakness that lets authenticated users with Duo Agent Platform access escape the prompt template sandbox through a specially crafted flow configuration, leading to remote code execution.

GitLab operates its own cloud-based AI Gateway used by GitLab.com, Self-Managed, and Dedicated customers, but users can also deploy self-hosted instances through GitLab Duo Self-Hosted. The company releases versions 19.2.4, 19.3.2, and 19.4.1 to fix the issue for self-hosted deployments, while customers using the GitLab-hosted gateway are already protected. GitLab reaches out directly to self-hosted customers and urges them to update immediately.

This warning follows GitLab's patching of a maximum severity path traversal vulnerability, CVE-2026-85706, in GitLab CE and EE last month, which lets unauthenticated attackers read sensitive data like credentials and secrets. CISA adds that flaw to its actively exploited list and gives federal agencies three days to secure their systems under BOD 26-04.

Read More at the original source →