Skip to content

Image: securityaffairs.com · rights & removal

GitLab Patches Max

Reporting by Security Affairs (Pierluigi Paganini)Read the original at securityaffairs.com

Objectivity score: 51 out of 100.
Difficulty 58 out of 100 (Graduate)— Read Analysis

Executive Summary

GitLab has released advisories regarding two security flaws. The first, CVE-2026-97229 (CVSS 10.0), involves an integer overflow bug in Community and Enterprise Editions affecting GitLab versions 19.2 before 19.2.8, 19.3 before 19.3.5, and 19.4 before 19.4.2, which allowed unauthenticated code execution via regular expression validation in CI/CD configurations. The second vulnerability, CVE-2026-90970 (CVSS 9.9), affects the GitLab AI Gateway, allowing an authenticated user with Duo Agent Platform access to execute arbitrary commands by escaping a prompt template sandbox via a specially crafted flow configuration. GitLab patched the AI Gateway issue by releasing versions 19.2.4, 19.3.2, and 19.4.1. The vulnerability in the AI Gateway did not require an unauthenticated connection.

Facts Only

* CVE-2026-97229 is an integer overflow bug from weak regular expression validation in CI/CD configurations.
* This flaw affects GitLab Community Edition and Enterprise Edition.
* Affected GitLab versions include 19.2 before 19.2.8, 19.3 before 19.3.5, and 19.4 before 19.4.2.
* The vulnerability allows an unauthenticated user to execute arbitrary code on the GitLab server via a specially crafted regular expression.
* CVE-2026-90970 is an issue in the GitLab AI Gateway.
* This flaw allows an authenticated user with Duo Agent Platform access to execute arbitrary commands on the gateway by escaping the prompt template sandbox through a flow configuration.
* GitLab released patches for CVE-2026-90970 in versions 19.2.4, 19.3.2, and 19.4.1.
* The AI Gateway vulnerability does not require an unauthenticated connection.
* The HackerOne researcher invisiblemeerkat reported CVE-2026-97229.

Full Take

The presence of two distinct vulnerabilities, one concerning general access control in CI/CD pipelines and another concerning privilege escalation within an AI gateway environment, points to systemic weaknesses across different layers of the platform. The juxtaposition of a zero-day-level risk (CVSS 10.0) against known, exploitable boundaries in an authenticated session (AI Gateway) suggests that security hygiene is inconsistent across core features and advanced modules. The fact that one vulnerability requires no authentication while the other relies on specific access means that defensive strategies must simultaneously address both public-facing configuration parsing and internal authorization checks. This creates a tension between securing public input mechanisms and securing internal, privileged operations. Who bears the cost of patching these disparate systems—the community users or those utilizing advanced AI features? How does the trust placed in an authenticated user translate into absolute security boundaries within specialized components like the AI Gateway?

From the original · Security Affairs (Pierluigi Paganini)

GitLab has addressed a critical security flaw, tracked as CVE-2026-97229 (CVSS score of 10.0), that could let attackers run code on its servers without logging in. The flaw affects Community Edition (CE) and Enterprise Edition (EE), it is an integer overflow bug caused by weak validation of regular expressions in CI/CD configurations.
Read the full story at securityaffairs.com

Sentinel — Human

Confidence

This text reads like a direct, factual report summarizing security advisories. It exhibits the structure and tone of technical journalism reporting on disclosed vulnerabilities.

Signals Detected
low severity: Sentence length variance is moderate; the text flows factually but lacks high rhetorical flourish.
low severity: The text maintains a direct, informational tone and successfully links two distinct vulnerability disclosures without excessive hedging or unsupported opinion.
low severity: The structure relies on direct quotation from advisories and attribution, typical of factual reporting.
low severity: The specific CVE numbers, version numbers, and names are presented factually, suggesting reliance on verifiable source material.
Human Indicators
The direct presentation of technical details (CVEs, version ranges) from official advisories suggests human curation of source material.
The final element acts as a standard media closing (follow links) rather than an internal statement.
GitLab Patches Max | Huntaegis