Posted in Vulnerability Management

Alpine Linux upgraded GitLab twice after the fix for a CVSS 10 flaw shipped. Their server was still…

Craig M. — Chief Product Officer @ Cyberr®

Alpine Linux upgraded GitLab twice after the fix for a CVSS 10 flaw shipped. Their server was still exploitable.

The story behind CVE-2026-97229, now on CISA's KEV list, is the most useful vulnerability management lesson I've read this year.

The bug lives in Onigmo, the regex engine inside Ruby. A crafted CI config could trigger an integer overflow and run code on the GitLab server. GitLab's fix was a patch to Ruby, applied only in its own packages and official images.

Alpine runs GitLab on its own image with standard Ruby. The version number said patched. The vulnerable regex engine never moved. An attacker registered a fresh account, crashed workers eight times, and even pulled Alpine's image to tune the exploit. Alpine found no evidence of code execution, then rebuilt from scratch, reset every password and replaced secrets anyway. Right call.

What I'd check this week:

→ Verify the component, not the banner. On custom builds, Ruby must be 3.3.13, 3.4.12, 4.0.9 or later.

→ Inventory every self-built image of a vendor product. That's where vendor-bundled patches quietly go missing. An SBOM per image turns this into a query, not a hunt.

→ Close open sign-up on self-managed GitLab or require admin approval. Step one of this attack was creating an account.

→ Treat worker crashes and 502 spikes on dev tooling as security signals, and keep logs beyond 14 days.

→ If in doubt, rotate CI variables and runner tokens. They're the real prize on a CI server.

* Patch notes tell you what the vendor shipped. They don't tell you what you're running. *

How are you validating fixes in images you build yourselves?

#VulnerabilityManagement #DevSecOps #SupplyChainSecurity

5 reactions · 1 comments

Topics: #devsecops, #vulnerabilitymanagement, #supplychainsecurity

Comments

Schomer: Great breakdown, Craig. The Alpine case is a brutal reminder that static configuration scanning often misses deep, multi-tier dependency chains. Intercepting malicious payloads (like crafted CI configs) at the ingestion layer rather than relying purely on runtime patching or version banners is critical. We've been analyzing this pattern with our ETBV Layer Zero pre-filter engine—specifically looki

Read and join the discussion on Cyberr