Winter CMS EUVD-2026-65532

Requires an authenticated Winter CMS backend account holding at least one of the CMS template-editing permissions cms.manage_pages, cms.manage_layouts, or cms.manage_partials - these are the exact prerequisites named in the advisory. The Twig sandbox escape then works against affected versions (>= 1.2.7, < 1.2.13) in their normal configuration; no additional non-default feature toggle is needed beyond having those permissions. Limiting factors: it is not exploitable by anonymous or front-end users, the required permissions are meant to be granted only to trusted developers/administrators (PR:H), and the CVSS records passive user interaction (UI:P) tied to the malicious template being rendered - though an attacker can trigger that rendering themselves.

The CVSS 4.0 base score is 9.3, driven by network reachability of the backend (AV:N/AC:L) and full confidentiality/integrity/availability impact including scope/subsequent-system compromise via RCE. The critical moderating signal is PR:H: exploitation requires an authenticated backend account already granted CMS template-editing permissions, which are intended for trusted developers/administrators - so this is fundamentally a privilege-boundary/defence-in-depth failure rather than an anonymous internet-facing RCE. There is no CISA KEV listing and no EPSS score in the provided data, so widespread automated exploitation cannot be asserted; the realistic threat model is a malicious insider, an over-provisioned low-trust content editor, or an attacker who has phished/compromised a backend credential and then pivots to full server and database takeover. Because a working patch and detailed payloads are public while active exploitation is unconfirmed, defenders should treat this as a high-priority patch for any multi-tenant or delegated-editing deployment, but a lower-urgency issue where template-editing is restricted to a handful of already-fully-trusted administrators.

An attacker who controls a backend account with cms.manage_pages (for example an over-privileged content editor or a phished developer login) edits a CMS page and inserts a Twig payload that forwards through an Eloquent model to the query builder - e.g. writing PHP into the page's own code section via a newQuery().update({ code: "<?php ... ?>" }) style call - turning the stored template into executable PHP. When the page is rendered the injected PHP runs on the server, and the same primitive can instead run arbitrary SQL to dump backend_users credentials. The advisory and shipped regression tests include concrete working payloads, lowering the effort required.