Executive brief
Flowise is a platform for building AI agents visually using document stores to manage embeddings and vector data. An authenticated attacker can exploit a flaw in the DocumentStore creation endpoint to overwrite or steal DocumentStore objects from other workspaces by submitting a request with an existing object's ID. This could allow an attacker to modify AI workflow configurations, redirect data indexing to their own infrastructure, or gain unauthorized access to embedding and vector storage settings belonging to other tenants in a shared environment.
Technical details
The vulnerability is a mass assignment and broken object-level authorization (BOLA) flaw in the DocumentStore entity creation logic. The root cause is that Flowise accepts a client-supplied primary key (id) without validation and passes the entire request body directly to the ORM's save() operation. TypeORM's save() method performs an UPDATE if the primary key already exists, converting the create endpoint into an implicit UPSERT operation. An authenticated attacker can obtain a valid DocumentStore UUID from another workspace and submit a POST /api/v1/document-store request with that ID, causing the existing record to be overwritten. Since the primary key is globally unique (not scoped to workspace) and the service layer does not validate ownership before updating, an attacker can effectively reassign a DocumentStore to their workspace. The vulnerability requires authentication and network access, but no special privileges. Flowise fixed this in version 3.1.0; all versions ≤ 3.0.13 are affected.
Affected products
- FlowiseAI flowise < 3.1.0
Timeline
- 2026-04-17: disclosed: Advisory GHSA-3prp-9gf7-4rxx published
- 2026-04-17: patched: Fixed in version 3.1.0