On 16 October 2026, security firm Rapid7 published a detailed technical analysis and working proof-of-concept (PoC) exploit code for a maximum-severity GitLab vulnerability, tracked as CVE-2026-97229. CVE stands for Common Vulnerabilities and Exposures, the public numbering system for security flaws. GitLab, a platform teams use to store code and run automated builds, fixed the flaw on 12 October. The company says that, under certain conditions, an unauthenticated person could run code on a server. Anyone running their own GitLab server should check their version today.
Exploit code arrives four days after the patch
GitLab released versions 19.4.2, 19.3.5 and 19.2.8 on 12 October and strongly recommended that every self-managed installation upgrade immediately. The flaw is an integer overflow, a maths error where a number outgrows the space reserved for it, in the part of GitLab that reads regular expressions (text-matching patterns) inside continuous integration and continuous delivery (CI/CD) settings. The Common Vulnerability Scoring System (CVSS) rating is 10.0, the highest possible. GitLab.com was already patched, and GitLab Dedicated tenants were fixed before the release. Rapid7 says its code runs commands on a server as GitLab’s own “git” user account.
Which GitLab versions need updating
GitLab says both Community Edition (CE) and Enterprise Edition (EE) are affected: every version from 19.2 before 19.2.8, from 19.3 before 19.3.5, and from 19.4 before 19.4.2. The same release fixed 24 vulnerabilities in total, one critical, five high, 14 medium and four low. GitLab also notes that 19.3.4 had a separate bug affecting merge request pages, so administrators on that branch should move straight to 19.3.5. Single-node installations will face downtime during the upgrade, because database migrations must finish before GitLab starts.
Is it really “no login needed”?
The headline claim needs care. GitLab’s score assumes no privileges are required. Rapid7’s analysis, however, describes one working route that begins after an attacker has already pushed a malicious CI/CD configuration and a specially named file path to a public project. Only the final request needs no login. Rapid7 says that prerequisite conflicts with GitLab’s rating and, because the original bug report is still private, it does not know how the reported attack avoids it.
GitLab’s release notes do not report any attacks. Rapid7 said on 13 October that it knew of no confirmed exploitation in the wild, a statement made before its full analysis appeared. Its code is tuned to one official CE 19.4.1 package on Linux, needs a little over 2 gigabytes of memory, and crashes a server process, leaving a “bad gateway” (502) error that defenders can look for in logs.
An AI-related bug fixed in the same release
The update also closes CVE-2026-96893, a high-severity flaw (7.3 out of 10) in GitLab Duo Agent Platform, which brings artificial intelligence (AI) agents into Enterprise Edition. GitLab says a signed-in user could slip crafted content into an agent’s instructions, a technique called prompt injection, causing commands that appear to come from other users. GitLab’s own staff found it. It is a separate flaw from the code-execution bug.
Our analysis: what this means
This section is our own reading, not a reported fact. Public code makes patching urgent, even if the exploit needs conditions many servers do not meet. The gap between GitLab’s “no privileges” score and Rapid7’s findings suggests real risk depends on who can create public projects on a given server. Open servers look more exposed than closed internal ones. GitLab’s previous maximum-severity flaw, CVE-2026-85706, was later added by the US Cybersecurity and Infrastructure Security Agency (CISA) to its Known Exploited Vulnerabilities (KEV) catalogue, according to BleepingComputer. That precedent argues for acting before attacks are confirmed. Rapid7’s temporary advice is to limit credential-free access to public projects, but it does not replace upgrading.
