Flatpak EUVD-2026-88554

Requires a victim to install a malicious Flatpak application; the attacker must get their app onto the target (via a Flatpak remote or sideloaded bundle). The vulnerable behavior is the export of an application's .desktop and D-Bus .service files to the host without filtering fields against an allowlist, which the fix in 1.18.4 introduces. Limiting factors: this is local, requires user interaction (app installation) rather than remote exploitation, affects Linux systems running Flatpak versions prior to 1.18.4, and the impact is denial of service / unintended host-service interaction rather than direct code execution or data theft.

This is a locally-exploitable, malicious-app-required flaw with no CVSS vector, EPSS score, KEV listing, or public POC provided in the input - all severity signals must be inferred from the advisory text. Exploitation is gated on a user installing a malicious Flatpak application, after which unfiltered .desktop and D-Bus .service files can cause denial of service and unintended interactions with host services. This is a genuine sandbox-boundary/integrity concern rather than a remote critical, but its real-world urgency is moderate: it is not the highest-impact issue in the 1.18.4 batch (the sibling CVE-2026-97024/97023 privileged file overwrite/deletion flaws are more severe), and there is no evidence of active exploitation or automatable mass abuse. Treat as a defense-in-depth hardening fix - patch as part of the coordinated 1.18.4 update but do not prioritize it above the file-overwrite CVEs. Note: because no NVD CVSS or CWE is published yet, the assessed vector below is independent analysis, not vendor-confirmed.

An attacker publishes a malicious Flatpak app (e.g. on a third-party remote or Flathub-style repository) that ships crafted .desktop and D-Bus .service files containing fields outside the intended allowlist. When a victim installs the app, Flatpak exports these files to the host, letting the app trigger denial of service or trigger unintended interactions with host services. No public exploit code is identified at time of analysis.