GitLab CE/EE EUVD-2026-25040

Requires authenticated user (typically with elevated privileges like Maintainer or Owner) to visit attacker-controlled website while maintaining an active GitLab session. The victim must have browser cookies enabled and not be using CSRF-resistant browser configurations. Attack vector is network-based (AV:N) but necessitates user interaction (UI:R) for successful exploitation - specifically, the victim browsing to a malicious third-party domain. No special GitLab configuration or non-default features required; the vulnerability exists in default installations. Attack complexity is low (AC:L) as CSRF payloads are straightforward to construct once GraphQL mutation structure is known. Exploitation is limited by the need for social engineering (phishing links, watering hole attacks) to deliver the malicious page to authenticated users, and by browser Same-Origin Policy which the CSRF protection should enforce but fails to. The attack does not work against users with expired sessions, those accessing GitLab exclusively via API tokens (bypassing browser cookie flow), or in environments with properly configured SameSite cookie policies at the infrastructure level.

CVSS 8.1 (High) reflects network attack vector with no privileges required but mandatory user interaction (AV:N/AC:L/PR:N/UI:R), enabling high confidentiality and integrity impact without availability damage (C:H/I:H/A:N). CISA SSVC framework categorizes this as non-automatable with total technical impact but no confirmed exploitation, suggesting lower priority than the CVSS score implies. The publicly available HackerOne exploit (report 3627285) demonstrates feasibility but requires social engineering to lure authenticated GitLab users to attacker-controlled websites. EPSS data not provided in intelligence feed. Absence from CISA KEV indicates no widespread exploitation campaigns detected. Real-world risk depends heavily on organizational exposure: internet-facing GitLab instances with administrative users are higher risk, while air-gapped or internal-only deployments face reduced threat. The requirement for user interaction (browsing to malicious site while authenticated to GitLab) significantly constrains mass exploitation compared to pre-auth RCE vulnerabilities, but targeted attacks against GitLab administrators remain viable, potentially enabling repository manipulation, CI/CD pipeline poisoning, or secret exfiltration through GraphQL mutations.

An attacker creates a malicious website containing hidden forms or JavaScript that issues GraphQL mutation requests to the victim's GitLab instance. When an authenticated GitLab administrator visits the attacker's site (via phishing email, compromised third-party site, or malicious advertisement), their browser automatically includes GitLab session cookies with the forged requests. The attacker's crafted mutations execute with the victim's privileges, potentially adding the attacker as a repository maintainer, modifying CI/CD pipeline configurations to inject malicious code, creating access tokens, or exfiltrating repository contents through GraphQL queries. The HackerOne report 3627285 provides working proof-of-concept code demonstrating the attack technique.