Summary
Flowise exposes OpenAI Assistants Vector Store CRUD endpoints without proper authorization checks, allowing any holder of an API key to perform unrestricted operations.
This becomes risky when API keys are broadly distributed or exposed, as attackers can read, modify, or delete vector store data without additional permission validation.
Affected context
The vulnerability occurs in Flowise deployments that expose the /api/v1/openai-assistants-vector-store endpoints.
These endpoints are not covered by the main authentication middleware when used with API key–based access and do not implement any fine-grained permission checks.
Environments where API keys are shared across services or users, or where keys are not tightly controlled, are particularly exposed to unauthorized vector store CRUD operations.
Exploit preconditions
- Attack surface: HTTP API endpoint
/api/v1/openai-assistants-vector-store - Required access: Possession of a valid Flowise API key; no additional role or permission is required
- Trigger point: Any CRUD request (create, read, update, delete) to the vector store endpoints using the API key
- Why exploitation is possible: The route is not whitelisted, yet it bypasses the main auth middleware and lacks per-operation authorization checks, so the system implicitly trusts any API key holder
Why this happens
The root cause is a missing authorization layer on a sensitive API route.
While the endpoint is nominally protected by API key authentication, there is no subsequent verification that the caller is authorized to perform vector store operations.
This aligns with CWE-862: Missing Authorization, where the application fails to enforce access control on a resource or function, leading to overbroad access for authenticated but unprivileged actors.
Potential impact
- Unauthenticated (beyond API key) users could read vector store contents and associated metadata
- Attackers with an API key could modify or delete existing vector store entries, impacting application behavior
- Malicious data could be injected into vector stores, influencing LLM responses and decision flows
- Integrity and confidentiality of ML pipelines and knowledge bases could be compromised
General mitigation guidance
Effective mitigation focuses on enforcing explicit authorization controls around sensitive API routes.
Recommended patterns include:
- Applying centralized auth middleware consistently to all API paths, including newly introduced routes
- Implementing role- or scope-based access checks for vector store operations, separating read and write capabilities
- Restricting API key privileges by default and associating each key with minimal required permissions
- Introducing defense-in-depth controls such as rate limiting, logging, and anomaly detection for critical data operations
- Regularly reviewing routing and middleware configurations to ensure no endpoints bypass intended policy enforcement
How teams detect issues like this
Detection typically involves a combination of configuration review and behavioral monitoring:
- Reviewing routing and middleware configurations to identify endpoints not protected by global auth or authorization layers
- Employing static analysis or policy-as-code checks to ensure all controller methods handling sensitive data invoke authorization logic
- Auditing access logs to spot unusual or high-volume CRUD activity on vector store endpoints, especially from generic or shared API keys
- Using security testing tools or manual penetration testing to systematically probe APIs for missing authorization on authenticated routes

