Technical impact
RAGFlow, an open‑source Retrieval‑Augmented Generation engine, contains a flaw that allows an authenticated user to invoke an OpenAI‑compatible completion endpoint for a private agent. This is a CWE-862 vulnerability, representing an authorization bypass through privileged attribute. The POST /api/v1/agents/chat/completions route returns the result from the openai‑compatible branch before the UserCanvasService.accessible() check runs, enabling the caller to read a private agent’s output even when the agent is configured as "Only me". The impact is the disclosure of private agent responses and, potentially, execution of configured tools that may have side effects, leading to confidentiality and integrity violations across tenant boundaries.
Affected Systems
The vulnerability affects Infiniflow RAGFlow versions from 0.25.2 up to and including 0.27.3. Any deployment using these versions is at risk unless updated.
Risk and Exploitability
The CVSS base score is 7.1, indicating high severity. EPSS information is not available, and the vulnerability is not listed in CISA's KEV catalog. The likely attack vector requires a legitimate authenticated user who knows the UUID of another user's private agent. The attack does not require elevation to administrative privileges, but it does allow an attacker to bypass authorization checks and access cross‑tenant data.