GitLab Enterprise Edition EUVD-2026-107672

Requires an authenticated user account on a GitLab EE (Enterprise Edition) instance running an affected version (17.8 up to the fixed 19.3.6/19.4.3/19.5.1 releases). Exploitation is limited to EE deployments that expose the paid AI features whose billing eligibility is checked server-side; Community Edition and instances not offering these AI features are not in scope. The limiting factors are the need for valid authentication (PR:L - not anonymous) and that the impact is confined to unauthorized access to subscription-gated AI functionality rather than access to other users' data or administrative control. The phrase 'under certain conditions' in the description implies the bypass is not universal across all request paths, but the exact conditions are not detailed in the available data.

This is a low-to-moderate priority issue that is best characterized as a billing/authorization bypass rather than a classic security breach. The provided CVSS 3.1 vector (AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:L/A:N, base 4.3 Medium) is internally consistent: exploitation is over the network with low complexity but requires an authenticated low-privileged account (PR:L), there is no confidentiality or availability impact, and only a low integrity impact (CWE-348, reliance on an untrusted/less-trusted source during billing eligibility checks). No EPSS score, CISA KEV listing, SSVC decision data, or public POC was provided, so there is no public exploit identified at time of analysis and no evidence of active exploitation. The real-world risk is primarily commercial/financial to GitLab (revenue leakage from unauthorized use of paid AI features) rather than a compromise of customer data or systems - the C:N/A:N scoring reflects that no victim data or service is directly harmed. Note the input Tags label this 'Information Disclosure,' which conflicts with the description and CWE-348: nothing is disclosed; an authenticated user improperly gains access to a paid feature, which is an access-control/business-logic flaw. Treat the Medium CVSS as an upper bound; absent exploitation signals this should be patched on a normal cycle, not emergency-prioritized.

An authenticated GitLab EE user on an instance without a valid paid AI subscription crafts or replays requests to AI feature endpoints whose server-side billing eligibility validation is insufficient, causing the backend to treat them as entitled. They then consume paid AI features (e.g., AI-assisted capabilities) without the subscription that should gate them. No public exploit code is known, and the low attack complexity means a logged-in user could trigger this without specialized tooling once the gap is understood.