GitLab's KEV'd CVSS 10 CI/CD Flaw Still Runs on More Than 1,000 Exposed Servers — and Alpine Linux Just Rebuilt Its Entire Instance After Attackers Came Close

GitLab's KEV'd CVSS 10 CI/CD Flaw Still Runs on More Than 1,000 Exposed Servers — and Alpine Linux Just Rebuilt Its Entire Instance After Attackers Came Close

GitLab's KEV'd CVSS 10 CI/CD Flaw Still Runs on More Than 1,000 Exposed Servers — and Alpine Linux Just Rebuilt Its Entire Instance After Attackers Came Close

Three weeks after GitLab patched a maximum-severity remote code execution flaw, more than 1,000 internet-exposed GitLab servers are still vulnerable to what is now confirmed, actively exploited territory. Tracked as CVE-2026-97229, the bug is an integer overflow in the CI/CD Regular Expression Compiler — the Onigmo library bundled in Ruby — and it lets an unauthenticated attacker run arbitrary code by submitting a specially crafted regular expression in a CI/CD pipeline configuration. CISA added it to its Known Exploited Vulnerabilities catalog on October 29 and gave federal agencies until Sunday, November 1 to secure their servers. Shadowserver reported over the weekend that it is tracking "over 1000 GitLab instances vulnerable to CVE-2026-97229" — out of roughly 39,000 GitLab instances it sees daily.

The patch itself shipped on October 12 in GitLab CE and EE 19.4.2, 19.3.5, and 19.2.8. "GitLab has remediated an issue that, under certain conditions, could have allowed an unauthenticated user to execute arbitrary code on the GitLab server via a specially crafted regular expression in a CI/CD configuration, due to insufficient regular expression validation," the company said at the time, adding that it "strongly recommend[s] that all self-managed GitLab installations be upgraded to one of these versions immediately." GitLab.com already ran the patched version and GitLab Dedicated customers needed no action — but self-managed installs, by definition the ones nobody else patches for you, are the ones now sitting on the KEV list.

The sharpest illustration of what is at stake is Alpine Linux. On October 20–21, attackers hit the project's GitLab instance — which hosts the "aports" build recipes that produce every Alpine software package, making it a high-value supply-chain target — with malicious CI configurations targeting CVE-2026-97229. The attempts crashed eight workers; four were "partially successful," though none seemed to go all the way. Alpine found no tampering in repositories, packages, or releases up to October 27, but it had lost the logs that could prove code execution never happened. "Because we cannot rule out code execution, and the evidence we would need to confirm it no longer exists, we took the harder route and rebuilt the GitLab instance from scratch," the team explained — resetting every password and publishing a new container image on October 27. The attackers had also downloaded Alpine's GitLab container image, a detail the project read as preparation for something bigger.

Alpine's post-mortem holds a lesson every self-managed administrator should copy onto a sticky note: the project first installed GitLab 19.3.4 instead of the patched 19.3.5, and a later upgrade to 19.3.6 still left the flaw open — because the fix required a patched Ruby (3.3.13) that GitLab bundles in its official packages but Alpine's custom container image did not include. If you run GitLab in a custom-built container or a nonstandard package, "upgraded" and "patched" are not the same thing. This is also GitLab's second Regular Expression Compiler RCE in two months, after CVE-2026-110264 — which Shadowserver says still exposes over 11,300 instances — and it follows September's KEV'd commits-API flaw CVE-2026-85706.

What Should You Do?

  1. Upgrade to 19.4.2, 19.3.5, or 19.2.8 or later — and verify the Ruby inside — if your GitLab runs in a custom container image like Alpine's did, confirm the image actually contains the patched Ruby 3.3.13, not just the newer GitLab version number.
  2. Check whether you are actually exposed — CI/CD with untrusted input (public projects, external contributors, or imported pipelines) is the realistic attack path; if users can submit pipeline configs, treat this as internet-facing even on an internal network.
  3. Inventory your exposure the way Shadowserver does — over 1,000 instances were still findable three weeks after the patch; if you cannot name which of your GitLab instances are internet-reachable, that is the first gap to close.
  4. If exploitation is plausible, follow Alpine's logic — missing logs turn "no evidence of compromise" into "no evidence"; preserve what you need to prove integrity, and if you cannot, rebuild from scratch and rotate every secret the instance held: passwords, 2FA secrets, CI/CD variables, and integration credentials.

The WAF Angle

An unauthenticated RCE in a service your developers push code through is the CI/CD equivalent of the upload-endpoint bugs we spend our lives blocking at the web tier — except GitLab is usually exempted from WAF coverage because it is "internal tooling," and the KEV listing is the reminder that internal tooling has internet-facing surfaces too. A WAF or reverse proxy in front of a GitLab instance can blunt the obvious mass-scanning traffic, but the payload here is a syntactically valid regular expression in a legitimate pipeline config — the same class of "looks like normal input, does abnormal things" request that authorization-aware tooling, not just pattern matching, has to catch. The practical web-security takeaway is the same one from September's CVSS 10 GitLab flaw and October's AI Gateway advisory: treat your source-control and CI/CD plane as production attack surface, patch it on KEV timelines, and never let a version number on a custom-built image stand in for verified remediation.

Sources