Grav CMS CVE-2026-69089

Exploitation requires the attacker to hold an authenticated Grav editor (or higher-privilege) account on the target installation - the traversal path is injected through editor-authored Markdown image syntax specifying the watermark parameter in page content. The Grav instance must have image watermark processing active and reachable through the content editing workflow; instances where the watermark feature is disabled are not exploitable via this path. File disclosure is bounded by the OS-level filesystem permissions of the PHP/web server process (e.g., www-data): only files readable by that process account can be exfiltrated, limiting access to configuration files, application secrets, and other assets within that permission scope. Once the malicious page is saved and rendered, the composited cache image becomes accessible at a public, unauthenticated URL - meaning any anonymous visitor can retrieve the disclosed file without additional attacker involvement. Sites with no external or untrusted editors, or those operating the server process under tightly restricted OS permissions, face substantially lower practical risk than multi-tenant Grav deployments.

The provided CVSS 4.0 vector assigns PR:N (no privileges required), yielding a score of 8.7, but this directly contradicts the CVE description, which states exploitation is performed by 'an editor authoring Markdown image syntax' - implying authenticated editor-level access is a prerequisite. This discrepancy is material: PR:N would imply unauthenticated remote file disclosure against any Grav instance, while the description constrains the attack to credentialed editors. The accurate privilege level is PR:L, which meaningfully reduces real-world exposure and should be independently verified against GHSA-w3f4-8pj2-599w. CVE-2026-69089 is not in CISA KEV and no public exploit code has been identified; EPSS data was not provided. The confidentiality impact is genuinely high - the traversal can reach any file readable by the PHP/web server process - but is bounded by OS-level filesystem permissions. Risk is highest for multi-tenant deployments with external or untrusted editors and lowest for single-author or small trusted-team installations where the editor role cannot be obtained by an adversary.

An attacker holding a Grav editor account composes a Markdown image directive embedding path traversal sequences in the watermark parameter - for example, targeting ../../config/security.yaml or other credential and configuration files readable by the web server process. When the page is saved and rendered, ImageMedium::watermark() processes the unsanitized path, reads the target file outside the media sandbox via the resource locator, and the image compositing layer embeds the file's binary content into a generated output image. The resulting cached file is then accessible at a predictable public URL, enabling any unauthenticated visitor to download and extract the disclosed content without further attacker interaction.