GitLab CVE-2026-110264 and CVE-2026-97229 score CVSS 10.0. An open sign-up page decides how easy they are to exploit.
GitLab has shipped two CVSS 10.0 "unauthenticated" code execution bugs in ten days, and the label is only half true. When Rapid7 rebuilt the first exploit, it found the attacker needs a way to plant a file in a public project. On any self-hosted GitLab that lets strangers sign up, a free account is enough.
.gitlab-ci.yml to a public project first. Open self-registration removes that hurdle.CVE-2026-110264 and CVE-2026-97229 are two maximum-severity flaws in the regular expression compiler of GitLab Community Edition and Enterprise Edition that can let an attacker run code on the GitLab server. A regular expression is a search pattern, like "any word ending in .com". GitLab compiles those patterns when it reads project files, including the .gitlab-ci.yml file that tells GitLab how to build and test code.
The bug is an integer overflow: a number gets so big it wraps around to a small one, like a car odometer rolling past 999,999. Feed GitLab a carefully built pattern and the compiler miscounts the memory it needs, which an attacker can turn into code execution.
Both were reported by the same researcher, joaxcar, through GitLab's HackerOne bug bounty.
Not quite: in Rapid7's reproduction of CVE-2026-97229, the attacker had to commit a malicious .gitlab-ci.yml and a file path containing UTF-8 characters to a branch in a public project before triggering the bug. Anyone with an ordinary account can do that if the server lets users create public projects. Rapid7's warning is the key line:
"On GitLab instances where new users can self register and create projects, this constraint is removed."
So the real question for each client isn't "is GitLab exposed?" It's "can a stranger make an account?" Open-source projects, agencies and schools often switch sign-up on deliberately, and a test server set up years ago may still have it on without anyone remembering.
GitLab hasn't said whether CVE-2026-110264 has the same requirement. Treat it as reachable until GitLab publishes details, which CSO Online notes it usually does 90 days after a fix.
No in-the-wild exploitation of CVE-2026-110264 or CVE-2026-97229 has been reported, and neither is on CISA's Known Exploited Vulnerabilities catalog as of 2026-10-22. That window may be short. Rapid7 reported reproducing code execution with CVE-2026-97229 on October 13, and GitLab servers are a proven target: GitLab's September flaw, CVE-2026-85706, was added to the KEV catalog on September 11, 2026.
Why do attackers love a GitLab server? Because it holds the keys. Rapid7's exploit runs commands as GitLab's git user, the account that owns every repository. From there an attacker can read client source code, CI/CD variables (where teams park cloud keys and deploy passwords) and runner tokens that reach production.
Upgrade every self-managed GitLab CE or EE instance to 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10 or 18.11.13, which fix both bugs. BleepingComputer confirms GitLab.com is already patched; GitLab Dedicated tenants are covered too. For each client:
git user. GitLab also lists two threat detections, "Onigmo native worker crash" and "Org mode render error burst", in its 19.5.1 advisory.A patch note tells you a bug exists. It doesn't tell you which of your clients left the door unlocked. That depends on settings, like open sign-up, public projects or a forgotten second instance, that live outside any patch dashboard.
KinoSec runs an autonomous pentest against each client's external surface: it finds the GitLab hosts, tells you which are reachable from the internet, checks whether a stranger can register, and reports what an outsider could do from there. Every finding comes with the remediation steps, in a report you can hand your client. Run it after the upgrade too, as proof the fix landed.
CVE-2026-110264 is a CVSS 10.0 integer overflow in GitLab CE/EE's regular expression compiler, fixed on October 21, 2026. GitLab says it could let an unauthenticated user run arbitrary code on the server under conditions it hasn't disclosed. It affects every self-managed version from 7.0 before the fixes.
First, load https://<their-gitlab>/users/sign_up from outside their network; a registration form means strangers can create accounts. Second, sign in as an admin and compare the version on the Admin dashboard against the fixed releases. Third, review the user list and runners for anything nobody remembers adding.
GitLab 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10 and 18.11.13 fix CVE-2026-110264 and include the CVE-2026-97229 fix. GitLab.com and GitLab Dedicated need no action.
Yes, for the part that depends on configuration. An autonomous pentest like KinoSec's can find internet-facing GitLab servers, confirm whether self-registration and public projects are open, and report the path an attacker would take. It runs with the client's permission and within agreed scope.