Loris CVE-2026-34985

This vulnerability presents moderate real-world risk with notable contextual factors. The CVSS score of 6.3 (Medium severity) with vector AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:L/A:L reflects that exploitation requires network access and low attack complexity, but authentication is mandatory (PR:L). The integrity and availability impacts are rated as 'Low' rather than 'High,' likely because the vulnerability enables read access and potentially minor file modification but does not enable system-wide compromise. However, the real-world risk is elevated because: (1) LORIS deployments handle sensitive neuroimaging research data protected by HIPAA, research protocols, and institutional review boards, making unauthorized access a compliance violation and research ethics breach, not merely a data exposure; (2) the attack requires knowledge of filenames but no complex exploitation technique (AC:L), making it practical if an attacker can enumerate file naming conventions; (3) the vulnerability affects all authenticated users in multi-project environments, potentially exposing data across unrelated research studies. The absence of KEV confirmation and public POC availability suggests limited active real-world exploitation at present, but the low barrier to exploitation (knowing a filename) means risk would escalate if weaponized. Organizations deploying LORIS in production should prioritize patching, particularly if managing multi-tenant or multi-project research data.

An authenticated researcher at Institution A who is assigned to Project X logs into the LORIS instance and observes the naming convention used for media files in their project (e.g., 'study_123_baseline_scan_T1.nii.gz'). The attacker then directly requests a file from another project by guessing or socially inferring a similar filename (e.g., 'study_456_baseline_scan_T1.nii.gz'), and because the backend media endpoint does not verify authorization, the request succeeds and the unauthorized file is served. No additional exploitation steps are required; the attack is a simple HTTP GET request to a known or guessable URL path, with no user interaction needed (UI:N).