Summary
Flowise contains an insecure direct object reference in its OpenAI Assistants integration, where authenticated requests can reference credentials from other workspaces by supplying an arbitrary credential UUID. If an attacker can access the affected Assistants endpoints, the missing workspace-ownership check may expose assistant metadata, file and vector store listings, and permit uploads into victim workspaces.
Affected context
The vulnerability occurs in Flowise deployments using the OpenAI Assistants integration when credential lookup accepts a caller-supplied UUID without confirming that the credential belongs to the caller's workspace. The affected endpoints may permit cross-workspace enumeration and resource operations when authentication is present but workspace-scoped authorization is not enforced.
Exploit preconditions
- Attack surface: Authenticated OpenAI Assistants integration endpoints
- Required access: A valid authenticated Flowise account or session
- Trigger point: Supplying a credential UUID associated with another workspace
- Why exploitation is possible: Credential resolution does not consistently verify workspace ownership before returning or using the referenced resource
Why this happens
The underlying bug is an object-level authorization failure. The application treats a client-provided credential identifier as sufficient authority to access the referenced object, rather than enforcing a relationship between the authenticated principal, the credential, and its workspace. This behavior maps to CWE-639: Authorization Bypass Through User-Controlled Key, commonly known as insecure direct object reference.
Potential impact
- Attackers may enumerate assistant metadata across workspaces
- Sensitive file and vector store listings could be exposed
- Files could be uploaded into another workspace
- OpenAI integration credentials or connected resources may be abused
- Workspace confidentiality and integrity could be compromised
- Follow-on data exposure or resource manipulation may occur
General mitigation guidance
Mitigation generally focuses on enforcing authorization at every object access and reducing trust in client-supplied identifiers.
Common defensive patterns include:
- Apply server-side workspace and tenant ownership checks before reading or modifying any credential or assistant resource
- Resolve authorization from the authenticated principal and server-maintained relationships rather than request parameters alone
- Use centralized, deny-by-default authorization middleware for object-level access decisions
- Return indistinguishable responses for unauthorized and nonexistent objects where appropriate
- Limit credential permissions and isolate integration resources according to least privilege
- Add regression tests covering cross-workspace reads, listings, uploads, and mutations
How teams detect issues like this
Organizations can identify similar authorization weaknesses through:
- Static analysis and code review searches for database lookups driven directly by request-provided UUIDs
- Review signals showing object retrieval without an adjacent tenant, workspace, owner, or role check
- Integration tests that replay valid object requests across separate workspaces
- API policy enforcement requiring explicit resource-scope validation on every read and write operation
- Monitoring for repeated access attempts involving identifiers outside the authenticated workspace
Automated security review tools may help flag direct object references, but authorization tests should validate actual tenant isolation.

