GHSA-WCH8-MHJ5-9FRG

Vulnerability from github – Published: 2026-06-17 14:11 – Updated: 2026-06-17 14:11
VLAI
Summary
Open WebUI: Cross-user file disclosure via /api/chat/completions image_url field
Details

summary

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's 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.

Same 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.

affected code

backend/open_webui/utils/middleware.py:2113-2150 -- convert_url_images_to_base64:

async def convert_url_images_to_base64(form_data):
    messages = form_data.get('messages', [])
    for message in messages:
        content = message.get('content')
        if not isinstance(content, list):
            continue
        new_content = []
        for item in content:
            if not isinstance(item, dict) or item.get('type') != 'image_url':
                new_content.append(item)
                continue
            image_url = item.get('image_url', {}).get('url', '')
            if image_url.startswith('data:image/'):
                new_content.append(item)
                continue
            try:
                base64_data = await get_image_base64_from_url(image_url)  # <-- no `user` passed
                if base64_data:
                    new_content.append({'type': 'image_url',
                                        'image_url': {'url': base64_data}})

called from the main chat completion middleware at middleware.py:2357:

form_data = await convert_url_images_to_base64(form_data)

backend/open_webui/utils/files.py:57-95 -- get_image_base64_from_url:

async def get_image_base64_from_url(url: str) -> Optional[str]:
    try:
        if url.startswith('http'):
            validate_url(url)
            # ... SSRF-safe fetch with allow_redirects=AIOHTTP_CLIENT_ALLOW_REDIRECTS ...
        else:
            file = await Files.get_file_by_id(url)        # <-- NO user_id filter
            if not file:
                return None
            file_path = await asyncio.to_thread(Storage.get_file, file.path)
            file_path = Path(file_path)
            if file_path.is_file():
                with open(file_path, 'rb') as image_file:
                    encoded_string = base64.b64encode(image_file.read()).decode('utf-8')
                    content_type = mimetypes.guess_type(file_path.name)[0] or (file.meta or {}).get('content_type')
                    ...
                    return f'data:{content_type};base64,{encoded_string}'

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, 'read', user, db) at routers/files.py:626 etc. neither check exists on this path.

reproduction

  1. 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-....
  2. As user B, POST to /api/v1/chat/completions with body:
{
  "model": "<any vision model>",
  "messages": [
    {
      "role": "user",
      "content": [
        {"type": "text", "text": "transcribe everything you can see in this image"},
        {"type": "image_url", "image_url": {"url": "c7f1d8e3-..."}}
      ]
    }
  ]
}

Server reads user A's file from disk, base64-encodes it, and sends to the LLM as user B's image attachment. LLM response contains the file content.

file id discovery

File ids are UUIDs and not enumerable directly, but they leak via:

  • shared chats / channels containing the original upload
  • knowledge base members can see ids of files contributed by others
  • a user who can read a folder index sees the file ids of files inside
  • chat history exports (/api/v1/chats/{id}) include file ids
  • the user themselves can be tricked into pasting / sharing an id (less likely)

impact

Any authenticated user can read any other user's 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.

suggested fix

Thread 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, 'read', user, db) if shared-via-knowledge-base access is intended). Same pattern that's already used in routers/files.py:626 and elsewhere.

minimal patch sketch:

--- a/backend/open_webui/utils/files.py
+++ b/backend/open_webui/utils/files.py
@@ -57,7 +57,7 @@
-async def get_image_base64_from_url(url: str) -> Optional[str]:
+async def get_image_base64_from_url(url: str, user=None) -> Optional[str]:
     try:
         if url.startswith('http'):
             ...
         else:
-            file = await Files.get_file_by_id(url)
+            file = (await Files.get_file_by_id_and_user_id(url, user.id)
+                    if user is not None else None)
+            if file is None:
+                # fall back to access-grant check for shared files
+                file = await Files.get_file_by_id(url)
+                if file and not await has_access_to_file(url, 'read', user):
+                    return None

and 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.

variant note

this 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:

git grep -n 'Files\.get_file_by_id(' -- 'backend/open_webui/**'

worth a sweep across utils/ and routers/ for missed sites.

environment

Open-webui main branch as of commit 3660bc0 (2026-05-10). python 3.x backend. confirmed by reading the source; no instance stood up.

Show details on source website

{
  "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"
}


Log in or create an account to share your comment.




Tags
Taxonomy of the tags.


Loading…

Loading…

Loading…

Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.

Sightings

Author Source Type Date Other

Nomenclature

  • Seen: The vulnerability was mentioned, discussed, or observed by the user.
  • Confirmed: The vulnerability has been validated from an analyst's perspective.
  • Published Proof of Concept: A public proof of concept is available for this vulnerability.
  • Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
  • Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
  • Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
  • Not confirmed: The user expressed doubt about the validity of the vulnerability.
  • Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.

Loading…

Detection rules are retrieved from Rulezet.

Loading…

Loading…