Summary
A mass assignment flaw in Flowise's assistant update endpoint lets authenticated users overwrite server-controlled fields such as workspaceId, createdDate, and updatedDate.
By manipulating workspaceId, an attacker can reassign assistants into arbitrary workspaces, undermining tenant isolation in multi-workspace deployments.
Affected context
This vulnerability affects Flowise deployments that:
- Expose the assistant update API to authenticated users
- Support multiple workspaces or tenants sharing the same backend
- Rely on server-side metadata (workspaceId, createdDate, updatedDate) to enforce ownership and isolation
- Accept JSON or request bodies where client-supplied fields are directly mapped onto assistant objects
It becomes especially risky in shared, multi-tenant environments where assistants from different workspaces coexist.
Exploit preconditions
- Attack surface: HTTP API endpoint for updating assistants
- Required access: Any authenticated account with permission to update its own assistant resources
- Trigger point: Sending an update request where server-controlled fields (such as workspaceId) are included and modified in the payload
- Why exploitation is possible: The server performs mass assignment without sufficient whitelisting, validation, or authorization checks, allowing user-controlled input to overwrite ownership-related fields
Why this happens
The core issue is a mass assignment pattern where incoming request bodies are automatically bound to backend models without restricting which fields may be modified.
Ownership and lifecycle fields such as workspaceId, createdDate, and updatedDate are treated as writable, even though they should be server-controlled.
This maps to CWE-915: Improperly Controlled Modification of Dynamically-Determined Object Attributes, combined with CWE-284/CWE-639 improper access control, because the application does not verify that the user is allowed to move assistants between workspaces.
Potential impact
- Attackers could reassign assistants into other users' or teams' workspaces
- Cross-tenant data exposure if assistants configured for one workspace appear in another
- Confusion or misrouting of large language model flows and automation tasks
- Potential compliance and privacy violations in hosted, multi-customer environments
- Indirect privilege escalation if workspace membership implies broader capabilities or integrations
General mitigation guidance
Effective mitigation focuses on constraining which model attributes clients can modify and enforcing strict ownership checks:
- Introduce explicit field whitelists or DTOs for update operations, excluding server-controlled properties such as workspaceId and timestamps
- Maintain workspaceId and other ownership markers as read-only on the server, deriving them from authenticated context rather than client input
- Enforce authorization rules that verify the caller's relationship to the target assistant and its workspace before applying any update
- Centralize access control logic to ensure consistent enforcement across all endpoints that manipulate assistant resources
- Design APIs so that multi-tenant boundaries are defined by authenticated context and server-side associations, not by client-supplied identifiers
How teams detect issues like this
Organizations can detect similar issues by:
- Performing static analysis or code review to flag patterns where entire request bodies are directly assigned (e.g., object spread, ORM update with raw input) to persistence models
- Searching for update endpoints that accept workspaceId, tenantId, ownerId, or similar ownership fields from clients
- Reviewing controller or service logic to confirm authorization checks for resource ownership and workspace membership are present and consistent
- Implementing automated tests that attempt to move resources between tenants or workspaces using another user's credentials
- Using API security scanners or automated review tools configured with rules for mass assignment and broken access control

