OFFSEC · Offensive Security

GitLab’s exploited CVSS 10 flaw is fixed in Ruby, and upgrading GitLab alone may not reach it

CISA listed CVE-2026-97229 as exploited on 29 October, a day after Alpine Linux described a week-long attempt against its own GitLab. Alpine upgraded twice and stayed vulnerable, because the fix is a patch to Ruby's regular-expression engine that GitLab ships only in its own packages.

The maximum-severity GitLab flaw that CISA added to its exploited list on 29 October is not fixed by upgrading GitLab alone [7][9]. CVE-2026-97229 sits in Onigmo, the regular-expression engine compiled into Ruby. GitLab’s fix is a patch to that engine, and GitLab applies it only in its own packages and container images [1][3]. Alpine Linux found this out under attack. An attacker probed its self-hosted instance for a week, the project upgraded GitLab twice, and on 26 October it found the server still vulnerable [1]. Any GitLab built from source, on a distribution’s Ruby or from a community container image needs a fixed interpreter as well as a fixed application [1][2].

What the data says

GitLab released 19.4.2, 19.3.5 and 19.2.8 on 12 October. It rated CVE-2026-97229 at CVSS 10.0, with no privileges or user interaction required, and described an unauthenticated user executing code on the server through a crafted regular expression in a CI/CD configuration [3]. The researcher joaxcar reported it through HackerOne [3][10]. Rapid7 reproduced code execution on an unpatched 19.4.1 image within a day and confirmed the same input failed against 19.4.2 [10]. On 13 October it updated its advisory with a link to a public proof-of-concept repository [5].

The preconditions are disputed. Rapid7 says its demonstrated route needs the ability to commit a malicious CI file and a specially named path to a public project, so GitLab’s no-privileges rating does not fully hold [5][6]. Where users can self-register and create projects, that constraint disappears [5]. The final request in the chain is unauthenticated [6]. The original report is private, so other routes may exist [6].

Rapid7 traces the bug to an integer overflow in Onigmo’s quantifier expansion, which lets the compiler accept a pattern it should reject; code then runs as the GitLab git user [5][6]. GitLab’s own checks on pattern length, repeat bounds and compiled size did not stop it [6]. On 23 October Ruby released 4.0.9, 3.4.12 and 3.3.13 with fixes for three Onigmo bugs behind four GitLab CVEs: a double free (CVE-2026-89078), two compilation-size overflows (CVE-2026-93577 and CVE-2026-97229) and an internal ID overflow (CVE-2026-110264) [2]. Ruby says only applications that compile patterns from attacker-controlled strings are affected. Its regex timeout does not protect against them [2]. Ruby has not requested its own CVE identifiers and asks scanners to track the issues under GitLab’s [2]. Ruby 3.2 is past end of life and will get no fix [8][12].

GitLab shipped a second CVSS 10.0 fix on 21 October. CVE-2026-110264 affects every version from 7.0 and is fixed in 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10 and 18.11.13 [4]. The release notes list a Ruby regexp patch in the Omnibus and cloud-native builds [4]. The 12 October releases do not address it [13]. GitLab reported no exploitation of CVE-2026-110264 [4][13].

Alpine’s report sets out the timeline. On 13 October it upgraded to 19.3.4 rather than the fixed 19.3.5, which it calls a mistake [1]. A newly registered account began probing on 20 October. On 21 October GitLab workers crashed eight times, and in four crashes the attacker got partway through the exploit [1]. Alpine moved to 19.3.6 on 22 October. That did not help: its image used standard Ruby, which lacked the patch [1]. After Alpine blocked the account on 23 October, the attacker returned twice and failed [1]. The attacker had also downloaded Alpine’s GitLab container image [1][7]. Alpine closed GitLab to outsiders on 27 October, rebuilt it, reset all passwords and replaced GitLab’s secret keys [1].

Alpine found no evidence the exploit succeeded but cannot rule it out. A successful run leaves no crash and no web-log trace, some logs from the attack day are gone, and outbound connections were not logged [1]. It verified repository commits against a developer’s clone and all 45 builders, and says signing keys were never stored on the GitLab server [1]. The Hacker News notes Alpine has not said whether runner tokens, CI/CD variables or integration credentials were replaced [8].

The KEV deadline is 1 November [7][9]. GBHackers reports that the entry also requires forensic triage under a CISA directive; The Hacker News mentions no directive [7][9]. CISA has not said what evidence it relied on, and The Hacker News cites Alpine’s incident as the exploitation evidence [7]. Ruby dates its fixes to 23 October, while The Hacker News says they shipped on 22 and 23 October [2][8]. Debian’s tracker still shows stable ruby3.3 at 3.3.8 and lists the 3.3.13 update as a pending packaging task [14][12].

What it means

Version-based vulnerability management breaks here. A scanner keyed to GitLab’s version would have reported Alpine’s 19.3.6 as fixed for both flaws [1][4]. Ruby’s advice to track the bugs under GitLab’s identifiers makes this worse, because those identifiers map to GitLab package versions, not interpreter builds [2]. Source installs, distribution packages and community images are all exposed to the gap. So is any organisation that patched on Singapore’s CSA alert of 14 October, which predates the Ruby release and gave GitLab versions only [11][2].

The asset matters more than the bug. Alpine lists what a successful attacker could have reached: password hashes, two-factor secrets, CI/CD variables, runner tokens, webhook and mirror credentials, and private repositories [1]. A source-control server sits upstream of every build it feeds. Alpine could state that its releases were unmodified only because it had independent builders and kept its signing keys elsewhere [1].

Detection is asymmetric. Failed attempts crash workers and produce 502 errors; a successful one may leave nothing in web logs [1][6]. GitLab has published six named detections for exploitation attempts [3]. The exposure is also not confined to GitLab. Any Ruby application that builds a regular expression from outside input carries the same engine bugs [2][12].

What to do

  • Upgrade self-managed GitLab to 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10 or 18.11.13. The 12 October releases leave CVE-2026-110264 open [4][13].
  • Check the interpreter on any GitLab not installed from the official Omnibus package or cloud-native images. It needs Ruby 3.3.13, 3.4.12, 4.0.9 or later. Re-pull Alpine’s 19.3.6 image if it was pulled before 27 October [1][2].
  • Do not infer a fix from a distribution’s package version alone. Confirm the vendor has shipped the Onigmo fixes or a backport [12][14].
  • Until patched, disable open self-registration, stop untrusted users creating or writing to public projects, and restrict anonymous access to them. Rapid7 also suggests blocking the CI lint endpoint. Treat these as temporary [5][6].
  • Hunt for long-running CI lint requests ending in 502 errors, Puma worker crashes with regexp traces, sudden worker memory growth and shell processes spawned by GitLab workers. Deploy GitLab’s published detections [3][6].
  • If any attempt is found, assume compromise. Rebuild, reset passwords, replace GitLab secret keys and rotate runner tokens, CI/CD variables, access tokens, and webhook, mirror and integration credentials [1][8].
  • Log outbound connections from the GitLab host, keep logs well beyond 14 days, and sign release tags so releases can be verified independently of GitLab [1].
  • Inventory other Ruby applications that compile user-supplied patterns, and retire Ruby 3.2 and earlier, which will not be fixed [2][12].

The offensive security view

This is a patch gap created by packaging rather than by code. The fix lived in a dependency only the vendor’s own builds carried [1][3]. One researcher produced four CVEs in the same engine in about two months, and the root fixes landed upstream eleven days after the first GitLab release [2][3]. The attacker in Alpine’s case downloaded the target’s own container image and kept returning after being blocked [1]. That is patient, build-specific work. Regular-expression compilers that accept user input are now proven attack surface, and the private report behind CVE-2026-97229 leaves room for further routes [6]. Defenders should audit what their builds actually contain, not what the version string says.

Sources

Every R3KONX article cites its primary material. 14 sources, in order of first citation. Links open the original publication.

  1. GitLab security incident: attempted exploitation of CVE-2026-97229 Alpine Linux · 2026-10-28
  2. Regexp compilation fixes related to GitLab's CVE-2026-89078, CVE-2026-93577, CVE-2026-97229 and CVE-2026-110264 Ruby (ruby-lang.org) · 2026-10-23
  3. GitLab Critical Patch Release: 19.4.2, 19.3.5, 19.2.8 GitLab · 2026-10-12
  4. 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 · 2026-10-21
  5. CVE-2026-97229: Critical Unauthenticated Remote Code Execution in GitLab Rapid7 · 2026-10-13
  6. Rapid7 Analysis: GitLab CI/CD Regex Integer Overflow (CVE-2026-97229) Rapid7 · 2026-10-16
  7. CISA Adds Actively Exploited GitLab RCE and Cisco Firewall DoS Flaws to KEV The Hacker News · 2026-10-30
  8. Attacker Tried to Exploit Alpine Linux's GitLab via CVSS 10.0 Flaw Its Ruby Left Open The Hacker News · 2026-10-29
  9. CISA Confirms Active Exploitation of GitLab Vulnerability CVE-2026-97229 GBHackers · 2026-10-30
  10. GitLab admins urged to patch critical RCE flaw (CVE-2026-97229) Help Net Security · 2026-10-13
  11. Critical Vulnerability in GitLab Community Edition and Enterprise Edition (AL-2026-141) Cyber Security Agency of Singapore · 2026-10-14
  12. Ruby 4.0.9, 3.4.12, and 3.3.13 Fix the Onigmo Bugs Behind Four GitLab CVEs lilting channel · 2026-10-26
  13. GitLab patches critical CVE-2026-110264 affecting versions back to 7.0 IT-Connect · 2026-10-22
  14. Debian Package Tracker: ruby3.3 Debian · accessed 2026-10-31

Researched and written by the R3KONX analysis desk from the cited primary material. The CISA alert and Known Exploited Vulnerabilities catalogue entry could not be retrieved, so the listing date, deadline and any directive reference are taken from press reports that quote them; those reports differ on whether a directive applies. CISA has not published the evidence behind the listing, and the only public account of exploitation activity is Alpine Linux's, which reports an attempt it could not confirm succeeded. Reports differ on whether the Ruby fixes shipped on 22 or 23 October. The vulnerability report behind CVE-2026-97229 is private, so the attack preconditions described here rest on Rapid7's reproduction. Corrections to event@r3konx.asia.

R3KONX 2027

Work in this domain? So does the programme

4–6 May 2027, World Trade Centre Kuala Lumpur. Curated talks, hands-on training and The BattleGrid.