---
title: "GitLab patched 37 flaws, ten in its AI agent tools, the same day gitlab.com carried a hacked Optimizely script"
description: "The November 16 batch fixes authorization bypasses in the Duo and MCP tools GitLab sells as guardrails for AI coding agents, and landed the same day gitlab.com served malware through a hacked Optimizely script. Results on December 1 will show whether the agent pitch and the security record can both hold."
publisher: "The Inference"
section: "Security"
published: 2026-11-18T19:14:59.952Z
modified: 2026-11-18T19:14:59.952Z
canonical: https://theinference.org/article/gitlab-patched-37-flaws-ten-in-its-ai-agent-tools-the-same-day-gitlab-com-carried-a-hacked-optimizely-script
language: en
keywords: "GitLab, Cybersecurity, Vulnerabilities, Supply Chain Attack, AI Coding"
---

# GitLab patched 37 flaws, ten in its AI agent tools, the same day gitlab.com carried a hacked Optimizely script

> The November 16 batch fixes authorization bypasses in the Duo and MCP tools GitLab sells as guardrails for AI coding agents, and landed the same day gitlab.com served malware through a hacked Optimizely script. Results on December 1 will show whether the agent pitch and the security record can both hold.

On November 16, 2026 [GitLab](https://theinference.org/markets/companies/gitlab) shipped patch versions 19.5.6, 19.4.8 and 19.3.9, closing 37 CVEs across its Community and Enterprise editions [1]. Ten of the 37 sit in the Duo and MCP components GitLab sells as the control plane for AI coding agents, counted from [the advisory's own table](https://cyberpress.org/gitlab-patches-37-security-bugs/) [1].

## The agent guardrails carry the bypasses

The sharpest: CVE-2026-123142, in which an OAuth token scoped to GitLab's MCP server, the protocol that lets AI agents operate the platform, could approve and merge merge requests [1]. CVE-2026-129253 let users bypass admin-configured AI tool governance, and CVE-2026-121948 is an authorization defect in Duo Code Review [1]. Behind them sit older-fashioned flaws: a CI/CD race condition (CVSS 8.2) that exposed protected pipeline variables, and a workspaces flaw (7.6) that leaked Kubernetes secrets across namespaces [1]. The batch follows October's exploited remote-code-execution bug and the Gentlemen ransomware crew's use of stolen GitLab credentials, [both covered here in October](https://theinference.org/article/gitlab-s-four-cves-got-fixed-twice-by-ruby-and-by-a-free-ai-agent).

## The same day, gitlab.com was the delivery channel

That Monday, researchers at Report URI traced malicious JavaScript served through Optimizely, the web testing platform, to visitors of 178 organizations, from [Zoom](https://theinference.org/markets/companies/zoom) and the BBC to HSBC and GitLab itself [2]. [BleepingComputer](https://www.bleepingcomputer.com/news/security/optimizely-supply-chain-attack-delivers-malware-to-cnn-fox-news-bbc-salesforce-sites/) independently confirmed gitlab.com among the sites where the script loaded [2].

Of 1,611 Optimizely scripts checked, 188 carried the kit, which swapped crypto wallet addresses, skimmed cards typed into a site's own payment forms and steered older iPhones toward the Coruna and DarkSword exploit kits [2][6]. Helme estimates the code loaded on roughly 34 million page visits, in a range of 6 to 150 million, with about $7,000 reaching the attacker [2][5].

## Only 24 of 178 could have blocked it

Optimizely, not GitLab, was breached: the file still came from Optimizely's own servers, and the company says it blocked the flaw and stripped the code the same day [3][4]. Only 24 of the 178 organizations had a content security policy strict enough to block it; 68 had none [3].

*Chart: **Only 24 of the 178 sites had a policy strict enough to block the script** Pie chart of content security policy status across the 178 organizations that loaded the malicious Optimizely script: 24 had a policy strict enough to block it, 68 had no policy, and 86 had a weaker one.*

*Content security policy status of the 178 organizations Report URI found serving the malicious Optimizely script, in organizations. Help Net Security reports 24 with a policy strict enough to block the script and 68 with none; the remaining 86 is the difference to the total. Source: Help Net Security, November 17, 2026. [3]*

## December 1 prices the risk

Fiscal Q3 closed October 31, before either event, so the quarter's revenue is already booked [7]. What is not booked is the verdict of the security reviewers who sign enterprise deals. GitLab enters the print with July-quarter revenue of $286 million, up 21.3 percent, its slowest growth in 12 reported quarters, on an operating loss of $57 million [8]. The call to watch for: any breakout of Duo and agent revenue, and whether management names the November advisory at all. The quarter closed on October 31; the security review of it opens on December 1 [8].

*Chart: **Growth has cooled from 33% to 21% while the agent pitch goes on sale** Line chart of GitLab's year-over-year revenue growth by quarter, from 32.5 percent in the October 2023 quarter to 21.3 percent in the July 2026 quarter.*

*Year-over-year revenue growth, percent, from quarterly figures in GitLab's SEC filings, October 2023 quarter through July 2026 quarter. The December 1, 2026, report covers the October 2026 quarter. [8]*

No public record yet shows a customer pausing a GitLab purchase over the November CVEs or the Optimizely list, and none may: a monthly patch cycle that catches its own agent bypasses is the product working. The risk runs the other way, into deal review timelines that never reach a press release.

## Takeaway

GitLab closed ten authorization flaws in the Duo and MCP tools it sells as guardrails for AI coding agents the same day its own site loaded a hacked Optimizely script; December 1 results will show whether enterprise buyers notice.

## Sources

1. [Cyber Press News, GitLab Patches 37 Security Flaws, Including High-Severity CI/CD and XSS Vulnerabilities, 17 November 2026](https://cyberpress.org/gitlab-patches-37-security-bugs/)
2. [BleepingComputer, Optimizely hack infected CNN, Fox News, BBC, Salesforce sites with malware, 16 November 2026](https://www.bleepingcomputer.com/news/security/optimizely-supply-chain-attack-delivers-malware-to-cnn-fox-news-bbc-salesforce-sites/)
3. [Help Net Security, Malware spread through Optimizely hits 178 organizations, including BBC, CNN and Zoom, 17 November 2026](https://www.helpnetsecurity.com/2026/11/17/optimizely-vulnerability-malicious-javascript/)
4. [Optimizely Service Status, Web Experimentation Security Incident, 17 November 2026](https://status.optimizely.com/)
5. [Newsweek, Cyber Attack Hits 178 Major Websites: Full List of Affected Organizations, 18 November 2026](https://www.newsweek.com/cyber-attack-websites-affected-12725863)
6. [The Next Web, Hackers used Optimizely to plant malware on Zoom, BBC, CNN and HSBC sites, 17 November 2026](https://thenextweb.com/news/optimizely-hack-malware-zoom-bbc-cnn-hsbc-178-sites)
7. [Business Wire, GitLab To Announce Third Quarter Fiscal 2027 Financial Results, 17 November 2026](https://www.businesswire.com/news/home/20261117901418/en/GitLab-To-Announce-Third-Quarter-Fiscal-2027-Financial-Results)
8. [GitLab Inc., quarterly figures from its SEC filings, retrieved 18 November 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001653482)


---

GitLab patched 37 flaws, ten in its AI agent tools, the same day gitlab.com carried a hacked Optimizely script — The Inference. Canonical: https://theinference.org/article/gitlab-patched-37-flaws-ten-in-its-ai-agent-tools-the-same-day-gitlab-com-carried-a-hacked-optimizely-script
