Executive brief
AnythingLLM is an application that converts documents into context for AI chatbots. A security flaw allows users with 'manager' or 'admin' privileges to delete files uploaded by any other user across the entire system, even in workspaces they are not authorized to access. This could lead to unauthorized data loss and disruption of service in multi-user environments.
Technical details
AnythingLLM (versions 1.11.1 through 1.14.0) contains an Insecure Direct Object Reference (IDOR) vulnerability in the 'embed-parsed-file' flow. While the application performs an ownership check when reading a file, the subsequent deletion logic is contained within 'finally' blocks that execute regardless of whether the ownership check passed. An attacker with 'manager' or 'admin' privileges can exploit this by sending a POST request to '/api/workspace/:slug/embed-parsed-file/:fileId' with an arbitrary integer 'fileId'. Because the deletion logic lacks ownership scoping (CWE-639), the server deletes the target file by its primary key even if it returns a 'File not found' error to the user. This allows for cross-tenant data deletion. The issue is fixed in version 1.14.1.
Affected products
- Mintplex-Labs anything-llm 1.11.1 to 1.14.0
Timeline
- 2026-06-14: advisory: GitHub Security Advisory GHSA-r872-gr59-vf5w published
- 2026-06-24: disclosed: CVE-2026-55611 published to NVD
- 2026-06-24: patched: Fix included in version 1.14.1