← ALL POSTS
Advisories·12 min read

GitLab flaw CVE-2026-97229 is being exploited: what to do

A score-10 flaw lets attackers with no account run code on self-managed GitLab servers, and CISA says it is exploited. What it means and what to do, in plain words.

Researched and drafted with AI assistance · reviewed by Prince

Developing story — last updated 30 October 2026.

In one sentence: if your organisation runs its own GitLab server, update it today to GitLab’s 21 October release (19.5.1, 19.4.3, 19.3.6 or 19.2.9) or newer, because attackers are using a flaw that lets someone with no account take over the server, and the earlier 12 October fix leaves a related flaw open.

That sentence packs in a lot of terms a newcomer may not know, so this post explains them one at a time. You don’t need any tools or prior experience to follow it. If you only want the actions, skip to Three things to do today.

What GitLab is, and who needs to care

GitLab is software that teams use to store and work on code. It also runs CI/CD, short for continuous integration and continuous delivery: automated jobs that build, test and ship software every time someone changes the code. The instructions for those jobs live in a configuration file inside each project.

GitLab comes in two editions. CE, the Community Edition, is the free open-source version; EE, the Enterprise Edition, is the paid version with extra features. Both are affected here.

There are also two ways to use it. You can use GitLab’s own hosted service, GitLab.com, or GitLab Dedicated, a hosted service GitLab runs for a single customer. Or you can run GitLab yourself on your own server, which GitLab calls “self-managed”. That difference decides whether you have anything to do. GitLab says GitLab.com is already running the patched version, and that all GitLab Dedicated tenants were patched before the fix was announced. This alert is for self-managed servers.

The flaw has the identifier CVE-2026-97229. CVE stands for Common Vulnerabilities and Exposures, the public catalogue that gives each known security flaw a single ID so that everyone talks about the same problem. We checked this ID on both the official CVE record and NVD, the National Vulnerability Database run by NIST, the US National Institute of Standards and Technology. GitLab itself issued the ID: it is a CNA, a CVE numbering authority, meaning an organisation allowed to assign IDs for its own products.

What kind of flaw it is

GitLab calls it an integer overflow in the CI/CD regular expression compiler. Those are two ideas, so let’s take them in turn.

A regular expression, often shortened to regex, is a small pattern that describes text, such as “any word that starts with release-”. CI/CD configuration can include these patterns so that a job runs only when, say, a branch name matches. Before a program can use a pattern, it compiles it: it turns the pattern into an internal form the computer can run quickly.

An integer overflow is a counting error. MITRE’s weakness catalogue (CWE-190) describes it as a calculation whose result grows past the largest number its storage can hold, so it wraps round to a very small or negative number. The program carries on as if the number were correct, and that can lead to it reserving too little memory and then writing past the end of it.

One everyday picture fits well. Think of an old car’s mileage counter with six digits. Drive one mile past 999,999 and it shows 000,000. Nothing broke loudly. The counter simply now tells a lie that anyone who trusts it will act on. In GitLab’s case, according to the Ruby project, the number that wraps round is the size of a compiled pattern.

That Ruby detail matters. GitLab is written in Ruby, a programming language, and Ruby uses a component called Onigmo to handle regular expressions. The Ruby project says GitLab found that specially crafted patterns passed to Ruby could corrupt memory while they were being compiled, reported the bugs, and patched the copy of Ruby that GitLab ships. Memory corruption is the bridge from a counting error to RCE, remote code execution, which means an attacker making the server run programs of their choosing from across the network.

GitLab’s own detection notes add one more piece of history. A closely related flaw, CVE-2026-93577, was fixed in GitLab 19.2.7, 19.3.3 and 19.4.1, and published on NVD on 24 September 2026. GitLab Security describes CVE-2026-97229 as a bypass of that fix, and says the 19.4.1 fix was incomplete. The Ruby project maps both CVEs to the same underlying Ruby bug.

What is affected

GitLab, the CVE record and NVD all give the same ranges for CVE-2026-97229, in both CE and EE:

Version lineAffectedFixed for this CVE inAlso fixed for the related CVE-2026-110264 in
19.2before 19.2.819.2.819.2.9
19.3before 19.3.519.3.519.3.6
19.4before 19.4.219.4.219.4.3

The CVE record lists everything outside those ranges as not affected by CVE-2026-97229. Read the last column before you relax, though.

On 21 October 2026 GitLab published a critical patch release for a second flaw, CVE-2026-110264. It is also an integer overflow when compiling a crafted regular expression, it also lets an unauthenticated user run code on the server, and it also scores 10.0. GitLab says it affects every version from 7.0 up to 18.11.13, 19.0.10, 19.1.9, 19.2.9, 19.3.6, 19.4.3 and 19.5.1. By those ranges, a server updated on 12 October to 19.2.8, 19.3.5 or 19.4.2 is safe from CVE-2026-97229 but still exposed to CVE-2026-110264. GitLab’s 21 October notes don’t spell that out; it is our reading of the version ranges GitLab published. The practical upshot is simple: aim for the 21 October versions or newer, not the 12 October ones.

How serious it is

Severity is usually expressed with CVSS, the Common Vulnerability Scoring System, which rates a flaw from 0 to 10. Each score comes with a vector, a short code recording how the attack works. GitLab scored CVE-2026-97229 at 10.0, critical, the highest possible. NVD displays GitLab’s score; its own analysis was still under way when we checked on 30 October.

The vector is CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H. Read left to right, in plain terms:

  • AV:N, network. The attacker can reach the flaw across a network, not just from the machine itself.
  • AC:L, low complexity. The attack doesn’t depend on luck or rare conditions.
  • PR:N, no privileges. No account is needed. GitLab calls the attacker “an unauthenticated user”.
  • UI:N, no user interaction. Nobody has to click anything.
  • S:C, scope changed. The damage isn’t confined to the flawed component; it can spread to other parts of the system.
  • C:H/I:H/A:H, high impact on confidentiality, integrity and availability: the attacker can read, change and break things.

Compare that with the earlier CVE-2026-93577, which GitLab scored 9.9 because the attacker needed to be logged in. The jump from “needs an account” to “needs no account” is what makes this one so urgent.

Is anyone using it?

Yes, according to CISA, the US Cybersecurity and Infrastructure Security Agency. On 29 October 2026 CISA added CVE-2026-97229 to its Known Exploited Vulnerabilities catalogue, usually called KEV. KEV is CISA’s list of flaws with reliable evidence of real-world attacks. It set a due date of 1 November 2026, the deadline for US federal agencies covered by its directive. That’s a three-day window, which tells you how CISA rates the risk. If you’re outside the US government the date isn’t binding on you, but it’s a sensible yardstick. KEV marks use in ransomware campaigns as “Unknown”.

GitLab’s 12 October release notes don’t say whether the flaw had been exploited. The CVE record carries a CISA assessment, timestamped 13 October, that marks exploitation as “active”, the attack as automatable and the technical impact as total. Automatable means an attacker could run it against many servers without hands-on effort. No source we read says when attacks began, who is behind them or how many servers have been hit.

DateEventSource
24 September 2026Related flaw CVE-2026-93577 published on NVD (fixed in 19.2.7, 19.3.3, 19.4.1)NVD
12 October 2026GitLab releases 19.4.2, 19.3.5 and 19.2.8, fixing CVE-2026-97229GitLab
13 October 2026CVE record published; CISA assessment on the record, timestamped this day, lists exploitation as activeCVE record
21 October 2026GitLab critical patch release fixes CVE-2026-110264GitLab, NVD
23 October 2026Ruby 4.0.9, 3.4.12 and 3.3.13 fix the underlying regex bugsRuby project
29 October 2026Added to CISA’s KEV catalogueCISA KEV, NVD
1 November 2026KEV due date for US federal agenciesCISA KEV

What an attacker gets, and what they need first

What they get, according to GitLab: the ability to run arbitrary code on the GitLab server. A GitLab server typically holds an organisation’s source code and the CI/CD jobs that build and ship its software, so control of it is serious.

What they need first: network access to your GitLab server and nothing more, going by the vector and GitLab’s description. GitLab adds that this applies “under certain conditions” but doesn’t say what those conditions are, so we won’t guess. A server reachable from the internet is the most exposed case.

How to fix it

GitLab lists no workaround. Updating is the fix.

  1. Find out which version you run. If you don’t manage the server yourself, ask whoever does.
  2. Update to the 21 October release for your version line, or newer: 19.2.9, 19.3.6, 19.4.3 or 19.5.1 (GitLab also released 19.1.9, 19.0.10 and 18.11.13 for older lines). These include the CVE-2026-97229 fix and the CVE-2026-110264 fix. GitLab says all self-managed installations should upgrade immediately.
  3. Plan for a short outage. In its 12 October notes GitLab warns that single-node instances (GitLab running on one server) will be down while database changes finish. Larger multi-node setups can upgrade without downtime by following GitLab’s zero-downtime procedure.
  4. If you have more than one GitLab server, start with any reachable from the internet. That order is our suggestion, not GitLab’s: those are the ones an attacker with no account can reach most easily.

A note for developers outside GitLab: the Ruby project says any Ruby application that builds regular expressions from text supplied by outsiders can hit the same bugs. It recommends Ruby 4.0.9, 3.4.12, 3.3.13 or later, and warns that Ruby’s regex time limit and pattern length limits don’t protect against them. CISA’s KEV notes likewise say other products may share the underlying Ruby/Onigmo issue.

How to tell whether you’re exposed, or already hit

Exposure is the easy part: any self-managed server in the affected ranges is exposed to CVE-2026-97229, and anything older than the 21 October versions is exposed to CVE-2026-110264.

Whether someone has already used it is harder. GitLab has published six threat detections for self-managed instances. A threat detection is a written rule describing what an attack attempt looks like in the server’s logs, the running records a server keeps of what it did. GitLab’s six are named for what they look for, such as crashes of the regex component, rejected oversized patterns, and bursts of errors from the feature that checks CI/CD configuration. They are documents for whoever looks after your logs; you don’t need to understand them yourself, but you should make sure that person has them.

GitLab Security’s notes on the crash detection carry two warnings worth repeating. A crash is a by-product of an attempt, not proof that the attack succeeded. And the reverse holds too: a successful attack can be silent, so finding nothing does not clear you. The notes point to GitLab’s web server error log, puma_stderr.log, as the main place to look for those crashes.

No source has published indicators of compromise, the specific clues such as network addresses or file fingerprints that mark a known attack. Beyond GitLab’s detections, the general practice for any server where an attacker may have run code is to look for things nobody on your team did: new user accounts or access tokens, changed CI/CD settings, and programs the server doesn’t normally run. That is our general guidance, not something the GitLab sources list. CISA’s KEV entry also asks US agencies to carry out forensic triage, a first structured look for signs of intrusion. If you find anything odd, or can’t tell, treat it as an incident and bring in someone experienced in incident response before you wipe or rebuild anything, because rebuilding destroys the evidence.

If you can’t patch yet

GitLab offers no workaround, so these steps only lower the risk while you arrange the update. None of them comes from GitLab.

  • Take the server off the open internet. Allow access only from your office network or a VPN (a virtual private network, an encrypted link into your organisation’s network). Because the attack needs only network access, this shrinks who can try. Cost: remote staff and outside collaborators lose easy access.
  • Watch the logs more closely, using GitLab’s detections. Cost: someone’s time, and it won’t catch a silent success.
  • As a last resort, switch the server off. CISA’s KEV entry says to stop using a product where mitigations are unavailable. Cost: your teams can’t work on code or ship software until it’s back.

What’s still unknown or disputed

  • When attacks started and how widespread they are. CISA added the flaw to KEV on 29 October, while the CISA assessment on the CVE record, timestamped 13 October, lists exploitation as active. The sources don’t explain the gap. No source gives dates, numbers or attacker details, and ransomware use is unknown.
  • The “certain conditions”. GitLab uses the phrase without saying what the conditions are.
  • Upgrading twice. GitLab’s 21 October notes don’t tell 12 October upgraders to upgrade again; we draw that from the published version ranges. Confirm with GitLab’s release notes if you’re unsure.
  • Scoring. The 10.0 score is GitLab’s. NVD had not finished its own analysis when we checked.

We’ll update this post if GitLab or CISA publishes more detail or indicators.

Three things to do today

  1. Find out whether your organisation runs its own GitLab server, and which version. If you only use GitLab.com or GitLab Dedicated, GitLab says it has already patched those for you.
  2. Update every self-managed server to 19.5.1, 19.4.3, 19.3.6 or 19.2.9, or newer, starting with any server reachable from the internet. If you updated on 12 October, check you’re on these newer versions too.
  3. Ask whether anyone has already been in. Give GitLab’s six threat detections to whoever looks after your logs, and check for accounts, tokens or settings nobody on your team created. If anything looks wrong, or you can’t tell, get incident response help before you rebuild.

Sources

  1. GitLab Patch Release: 19.4.2, 19.3.5, 19.2.8 — GitLab · accessed
  2. GitLab Critical Patch Release: 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10, 18.11.13 — GitLab · accessed
  3. CVE-2026-97229 — NIST NVD · accessed
  4. CVE-2026-97229 (CVE record) — CVE Program (GitLab as CNA) · accessed
  5. Known Exploited Vulnerabilities catalog: CVE-2026-97229 — CISA · accessed
  6. Known Exploited Vulnerabilities catalog (JSON feed, version 2026.10.29) — CISA · accessed
  7. Regexp compilation fixes related to GitLab's CVE-2026-89078, CVE-2026-93577, CVE-2026-97229 and CVE-2026-110264 — Ruby project · accessed
  8. Detection: GitLab org-ruby or Onigmo native worker crash — GitLab Security · accessed
  9. CVE-2026-93577 — NIST NVD · accessed
  10. CVE-2026-110264 — NIST NVD · accessed
  11. CWE-190: Integer Overflow or Wraparound — MITRE CWE · accessed
  • gitlab
  • cve-2026-97229
  • known-exploited
  • remote-code-execution
  • ci-cd

Keep reading

Advisories10 min read

FreeSWITCH CVE-2026-116779: one SIP message can run commands, and it is exploited

Advisories6 min read

Brotli flaw CVE-2026-111497 is exploited: update Node.js, .NET and more

Advisories11 min read

SiteGuard Server Edition flaw CVE-2026-106931 is being exploited: what to do