{
  "affected": [
    {
      "database_specific": {
        "last_known_affected_version_range": "\u003c= 0.9.5"
      },
      "package": {
        "ecosystem": "PyPI",
        "name": "open-webui"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0"
            },
            {
              "fixed": "0.9.6"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-54009"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-639"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-06-17T14:11:44Z",
    "nvd_published_at": null,
    "severity": "MODERATE"
  },
  "details": "## summary\n\n`POST /api/chat/completions` accepts an `image_url.url` value that, when it does NOT start with `http://`, `https://`, or `data:image/`, is interpreted as a file id and resolved against the global file table with no ownership check. An authenticated user can therefore set `image_url.url` to another user\u0027s file id, the server reads that file from disk, base64-encodes it, and injects the data URI into the LLM request. The user then prompts the LLM to describe / OCR the file and reads the content back.\n\nSame class as CVE-2026-44560 (RAG cross-user access) and the multiple `has_access_to_file` checks added in `routers/files.py` -- the auth boundary was tightened on the file router but not on this conversion path.\n\n## affected code\n\n`backend/open_webui/utils/middleware.py:2113-2150` -- `convert_url_images_to_base64`:\n\n```python\nasync def convert_url_images_to_base64(form_data):\n    messages = form_data.get(\u0027messages\u0027, [])\n    for message in messages:\n        content = message.get(\u0027content\u0027)\n        if not isinstance(content, list):\n            continue\n        new_content = []\n        for item in content:\n            if not isinstance(item, dict) or item.get(\u0027type\u0027) != \u0027image_url\u0027:\n                new_content.append(item)\n                continue\n            image_url = item.get(\u0027image_url\u0027, {}).get(\u0027url\u0027, \u0027\u0027)\n            if image_url.startswith(\u0027data:image/\u0027):\n                new_content.append(item)\n                continue\n            try:\n                base64_data = await get_image_base64_from_url(image_url)  # \u003c-- no `user` passed\n                if base64_data:\n                    new_content.append({\u0027type\u0027: \u0027image_url\u0027,\n                                        \u0027image_url\u0027: {\u0027url\u0027: base64_data}})\n```\n\ncalled from the main chat completion middleware at `middleware.py:2357`:\n\n```python\nform_data = await convert_url_images_to_base64(form_data)\n```\n\n`backend/open_webui/utils/files.py:57-95` -- `get_image_base64_from_url`:\n\n```python\nasync def get_image_base64_from_url(url: str) -\u003e Optional[str]:\n    try:\n        if url.startswith(\u0027http\u0027):\n            validate_url(url)\n            # ... SSRF-safe fetch with allow_redirects=AIOHTTP_CLIENT_ALLOW_REDIRECTS ...\n        else:\n            file = await Files.get_file_by_id(url)        # \u003c-- NO user_id filter\n            if not file:\n                return None\n            file_path = await asyncio.to_thread(Storage.get_file, file.path)\n            file_path = Path(file_path)\n            if file_path.is_file():\n                with open(file_path, \u0027rb\u0027) as image_file:\n                    encoded_string = base64.b64encode(image_file.read()).decode(\u0027utf-8\u0027)\n                    content_type = mimetypes.guess_type(file_path.name)[0] or (file.meta or {}).get(\u0027content_type\u0027)\n                    ...\n                    return f\u0027data:{content_type};base64,{encoded_string}\u0027\n```\n\n`Files.get_file_by_id` in `models/files.py:161` does a bare `db.get(File, id)` -- no ownership filter. there is a separate `Files.get_file_by_id_and_user_id` at line 172 that does filter on `user_id`, and the file router uses `has_access_to_file(id, \u0027read\u0027, user, db)` at `routers/files.py:626` etc. neither check exists on this path.\n\n## reproduction\n\n1. As user A, upload any file (image works cleanly, pdf works if a vision-capable model is configured). Note the file id from the upload response, e.g. `c7f1d8e3-...`.\n2. As user B, POST to `/api/v1/chat/completions` with body:\n\n```json\n{\n  \"model\": \"\u003cany vision model\u003e\",\n  \"messages\": [\n    {\n      \"role\": \"user\",\n      \"content\": [\n        {\"type\": \"text\", \"text\": \"transcribe everything you can see in this image\"},\n        {\"type\": \"image_url\", \"image_url\": {\"url\": \"c7f1d8e3-...\"}}\n      ]\n    }\n  ]\n}\n```\n\nServer reads user A\u0027s file from disk, base64-encodes it, and sends to the LLM as user B\u0027s image attachment. LLM response contains the file content.\n\n## file id discovery\n\nFile ids are UUIDs and not enumerable directly, but they leak via:\n\n- shared chats / channels containing the original upload\n- knowledge base members can see ids of files contributed by others\n- a user who can read a folder index sees the file ids of files inside\n- chat history exports (`/api/v1/chats/{id}`) include file ids\n- the user themselves can be tricked into pasting / sharing an id (less likely)\n\n## impact\n\nAny authenticated user can read any other user\u0027s file content (image and any file with an image-guess mimetype path) via this channel. Severity is bounded by what the LLM will accept in `image_url` -- in practice, image files work cleanly with any vision model; pdf / docx work with multi-modal providers that accept them.\n\n## suggested fix\n\nThread the authenticated user through to `get_image_base64_from_url` and resolve the file via `Files.get_file_by_id_and_user_id(id, user.id)` (or `has_access_to_file(id, \u0027read\u0027, user, db)` if shared-via-knowledge-base access is intended). Same pattern that\u0027s already used in `routers/files.py:626` and elsewhere.\n\nminimal patch sketch:\n\n```diff\n--- a/backend/open_webui/utils/files.py\n+++ b/backend/open_webui/utils/files.py\n@@ -57,7 +57,7 @@\n-async def get_image_base64_from_url(url: str) -\u003e Optional[str]:\n+async def get_image_base64_from_url(url: str, user=None) -\u003e Optional[str]:\n     try:\n         if url.startswith(\u0027http\u0027):\n             ...\n         else:\n-            file = await Files.get_file_by_id(url)\n+            file = (await Files.get_file_by_id_and_user_id(url, user.id)\n+                    if user is not None else None)\n+            if file is None:\n+                # fall back to access-grant check for shared files\n+                file = await Files.get_file_by_id(url)\n+                if file and not await has_access_to_file(url, \u0027read\u0027, user):\n+                    return None\n```\n\nand pipe `user` through `convert_url_images_to_base64(form_data, user)` from the middleware caller. happy to send a PR once you confirm the fix shape you want.\n\n## variant note\n\nthis was found via patch-diffing existing advisories. the same bug class likely exists in any other site that calls `Files.get_file_by_id` without an adjacent `has_access_to_file` / `get_file_by_id_and_user_id` check. quick grep:\n\n```\ngit grep -n \u0027Files\\.get_file_by_id(\u0027 -- \u0027backend/open_webui/**\u0027\n```\n\nworth a sweep across utils/ and routers/ for missed sites.\n\n## environment\n\nOpen-webui main branch as of commit `3660bc0` (2026-05-10). python 3.x backend. confirmed by reading the source; no instance stood up.",
  "id": "GHSA-wch8-mhj5-9frg",
  "modified": "2026-06-17T14:11:44Z",
  "published": "2026-06-17T14:11:44Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/open-webui/open-webui/security/advisories/GHSA-wch8-mhj5-9frg"
    },
    {
      "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:H/I:N/A:N",
      "type": "CVSS_V3"
    }
  ],
  "summary": "Open WebUI: Cross-user file disclosure via /api/chat/completions image_url field"
}