GitLab patches AI Gateway flaw that could run commands on self-managed instances
The fix closes a route to command execution in self-hosted AI Gateway setups, while GitLab-hosted customers were already protected.
GitLab has patched two serious self-managed security issues that matter to teams running their own instances: one in the AI Gateway that could lead to remote command execution, and one path-traversal bug in CE/EE that was already being probed in the wild. For the AI Gateway flaw, tracked as CVE-2026-90970, GitLab says an authenticated user with Duo Agent Platform access could escape a prompt template sandbox and run commands on the gateway. It shipped fixes in 19.2.4, 19.3.2, and 19.4.1 for Self-Hosted AI Gateway users, while GitLab-hosted AI Gateway customers were already protected. The second issue, CVE-2026-85706, affects self-managed GitLab CE/EE, carries a CVSS score of 10.0, and can expose arbitrary files such as secrets and CI/CD data; GitLab says it was patched on September 11 and later added to CISA’s Known Exploited Vulnerabilities catalog.
Why it matters
For teams running their own GitLab AI Gateway, the patch removes a path that could let an authenticated user turn gateway access into command execution. That lowers the risk of a takeover on exposed self-managed instances and shifts attention to updating to the fixed releases before the flaw can be used against them.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.