GitLab urges self-managed users to upgrade after critical security fix

GitLab’s hosted service is already patched, but customers running their own installations face an urgent upgrade recommendation. On October 21, 2026, GitLab released security updates across seven release lines, including a fix for a flaw that, under certain conditions, could let someone run code on a server without logging in.

A crafted search pattern can reach server execution

The most severe issue, CVE-2026-110264, sits in the compiler for regular expressions—patterns used to match text. GitLab says compiling a specially crafted pattern could trigger an integer overflow, a numeric calculation error, and allow an unauthenticated user to execute arbitrary code. GitLab assigned it a CVSS severity score of 10.0.

The affected versions stretch back to GitLab 7.0, with separate fixed-version thresholds for the newer branches. Both Community Edition and Enterprise Edition are affected. Beyond its standard patch releases, GitLab backported this particular fix to 18.11.13, 19.0.10, 19.1.9 and 19.2.9.

For self-managed customers, GitLab also released two threat detections to help identify exploitation attempts: “Onigmo native worker crash” and “Org mode render error burst.” The company credits security researcher joaxcar with reporting the vulnerability through its HackerOne bug bounty program.

Seven patched versions, different vulnerability coverage

The release covers these Community Edition and Enterprise Edition versions:

• 19.5.1, 19.4.3 and 19.3.6
• 19.2.9, 19.1.9 and 19.0.10
• 18.11.13

The older-branch backports should not be read as fixes for every vulnerability in the announcement. GitLab lists the other security issues as fixed in 19.3.6, 19.4.3 and 19.5.1, with affected starting versions varying by flaw. Its recommendation is to move affected installations to the latest version as soon as possible.

GitLab also says that when an advisory does not name a specific deployment type, all deployment types are affected. Its policy is to make the issue-tracker details for security vulnerabilities public 90 days after the release that patches them.

Identity and AI controls also needed repairs

Two high-severity fixes address identity systems. One could let an authenticated user take over an existing LDAP-backed account when OmniAuth SAML and LDAP auto-linking were both enabled, without the victim’s interaction. Another could let an unauthenticated user poison a shared OAuth/OIDC discovery cache, redirecting certain authentication flows to an externally controlled server.

The security fixes include two narrower AI-related failures in Enterprise Edition. CVE-2026-112378 could let an authenticated user access paid AI features without a valid subscription because server-side billing eligibility checks were insufficient. GitLab rated that issue 4.3 and says it was discovered internally.

CVE-2026-103515 concerns an administrator’s lock on a group’s Duo availability. A user with group Owner permissions could, under the affected conditions, re-enable AI-backed data transfers to a third-party service for that group and its descendant projects. GitLab rated the flaw 2.7; it involved insufficient enforcement of the administrator’s lock for Amazon Q settings.

Other repairs close access-control gaps beyond AI: users could read milestone metadata from private projects without membership, bypass required merge-request approval policies under certain conditions, or permanently delete existing vulnerability records despite having read-only permissions. Each reflects a different permission boundary addressed in the same release.