Executive brief
LobeChat, an open-source AI chat framework, contains a security flaw in how it handles private documents uploaded by users for AI analysis. An authenticated user can bypass security restrictions to search through and read the contents of documents belonging to other users on the same server. This could lead to the exposure of sensitive information, such as private notes, API keys, or confidential business data stored within the application's knowledge base.
Technical details
A broken access control vulnerability (IDOR) exists in LobeChat's Retrieval-Augmented Generation (RAG) component. The `semanticSearch` and `semanticSearchForChat` methods in the `ChunkModel` failed to include a `userId` predicate in their database queries. Consequently, an authenticated attacker can perform vector similarity searches against the entire database corpus or specific `fileIds`/`knowledgeIds` belonging to other users. This allows for the retrieval of plain-text chunks, filenames, and metadata from documents the attacker does not own. The issue was addressed by enforcing ownership checks in the `ChunkModel` and scoping knowledge-base file resolution to the caller's `userId`.
Affected products
- LobeHub LobeChat through 2.2.9
Timeline
- 2026-06-04: disclosed: Initial researcher report to vendor
- 2026-07-01: patched: Fix merged in GitHub pull request #16594
- 2026-07-02: advisory: CVE published and NVD record created