GitLab CVE-2026-105777

Exploitation is gated by the vulnerable version ranges (17.6.5 before 19.3.6, 19.4 before 19.4.3, 19.5 before 19.5.1) and by specific 'certain conditions' under which GitLab improperly filters internal response headers, causing sensitive internal server credentials to be included in responses returned to the client; the precise triggering configuration/endpoint is not detailed in the available data and should be confirmed from the GitLab work item (gitlab.com/gitlab-org/gitlab/-/work_items/597022). The key limiting factor is the disputed privilege requirement: the CVSS vector indicates PR:L (an authenticated low-privilege session is needed), while the description claims an unauthenticated user suffices - this prerequisite is unresolved and materially affects realistic risk. No user interaction is required (UI:N).

This is a moderate-severity information-disclosure issue (CWE-201) with a provided CVSS 3.1 base score of 4.3 (AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:N/A:N). There is a clear CONFLICT between signals: the CVSS vector encodes PR:L (requiring some authenticated privilege), while the description explicitly states 'an unauthenticated user could have retrieved sensitive internal server credentials' - these cannot both be true, so the authentication prerequisite should be verified with GitLab before prioritization. If the description is accurate (unauthenticated, PR:N), real-world risk is meaningfully higher than the 4.3 score suggests, because leaked 'internal server credentials' can enable follow-on lateral movement. Publicly available exploit code exists (a HackerOne report is referenced), which raises exploitability, but there is no EPSS score, no SSVC decision data, and no CISA KEV listing provided, so there is no evidence of active exploitation - POC availability must not be conflated with in-the-wild use. Net assessment: a real but second-tier priority - patch promptly given GitLab's exposure and public POC, but it is not an emergency unless the unauthenticated-access claim is confirmed.

Against a vulnerable self-managed or hosted GitLab instance (17.6.5 before 19.3.6, 19.4 before 19.4.3, or 19.5 before 19.5.1), an attacker sends crafted requests under the specific conditions that trigger improper filtering of internal response headers and reads sensitive internal server credentials leaked back in the response. Publicly available exploit material is referenced via a HackerOne report (hackerone.com/reports/3657282), lowering the barrier to reproduction. Because the CVSS vector is AV:N/AC:L, the request can be delivered remotely over the network with low complexity, though the privilege level required is disputed between the vector (PR:L) and the description (unauthenticated).