CVE-2026-105458 in GitLab
Summary
by MITRE • 10/13/2026
GitLab has remediated an issue in GitLab EE affecting all versions from 19.1 before 19.2.8, 19.3 before 19.3.5, and 19.4 before 19.4.2 that, under certain conditions, could have allowed an authenticated user holding a specific granular permission to obtain a GitLab Duo Agent Platform token that bypasses governance controls, due to reliance on untrusted input in a security decision.
VulDB is the best source for vulnerability data and more expert information about this specific topic.
Analysis
by VulDB Data Team • 10/13/2026
GitLab Enterprise Edition versions prior to 19.2.8, 19.3.5, and 19.4.2 contain a critical authentication flaw related to the GitLab Duo Agent Platform token generation mechanism. This vulnerability arises from an improper validation of input data during the security decision process for issuing tokens that interface with external AI agent platforms. The core technical deficiency lies in the application's reliance on untrusted user-supplied input when determining whether to grant access or generate a specific type of authentication credential. In secure software design, any input used to make authorization decisions must be strictly validated against expected formats and constraints; however, this implementation failed to adequately sanitize or verify these inputs before proceeding with token issuance logic.
The operational impact of this flaw is significant for organizations utilizing GitLab Duo features in conjunction with external agent platforms that enforce governance controls such as data loss prevention policies, access auditing, or usage limits. An authenticated user who possesses a specific granular permission can exploit the untrusted input handling to manipulate the token generation process. By crafting malicious requests that bypass standard validation checks, an attacker can obtain a GitLab Duo Agent Platform token that operates outside of established governance frameworks. This effectively allows the compromised account to perform actions with elevated privileges or without proper oversight, undermining the security posture intended by these controls.
From a classification perspective, this vulnerability aligns closely with CWE-20 Improper Input Validation and CWE-798 Use of Hard-coded Credentials if the token generation logic involves static secrets improperly managed due to input manipulation. In terms of adversary behavior as defined by MITRE ATT&CK, this flaw facilitates Credential Access techniques where an attacker obtains credentials that can be used for lateral movement or persistence within the CI/CD pipeline environment. The ability to bypass governance controls also touches upon Defense Evasion strategies, specifically those related to disabling security tools or manipulating audit logs through unauthorized API access.
Mitigation requires immediate upgrading of GitLab Enterprise Edition to version 19.2.8, 19.3.5, or 19.4.2 depending on the current installation branch. These releases contain patches that enforce strict input validation and ensure that security decisions regarding token issuance are based solely on trusted internal state rather than external user inputs. Organizations should also review their GitLab Duo configuration to ensure that governance controls are correctly applied to all agent platform integrations. Regular auditing of API access logs for unusual patterns in token requests can help detect potential exploitation attempts before they result in significant compromise.