GitHub Advisoryadvisory · PyPI
GHSA-wch8-mhj5-9frg Open WebUI: Cross-user file disclosure via /api/chat/completions image_url field
MEDIUM
Official page ↗
- Aliases
- CVE-2026-54009
- Ecosystem
- PyPI
- Packages
- open-webui
- Severity
- Moderate
- CVSS vector
- CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N
- Weaknesses
- CWE-639
- Reviewed
- 2026-06-17
Affected packages
• PyPI open-webui: < 0.9.6 → fixed in 0.9.6
Description
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`:
```python
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) # 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) # ",
"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:
```diff
--- 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.
Related
Alias · CVE (1)
- CVE-2026-54009 Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. vulnerability
Weakness · CWE (1)
- CWE-639 Authorization Bypass Through User-Controlled Key weakness
References
- https://github.com/open-webui/open-webui/security/advisories/GHSA-wch8-mhj5-9frg web
- https://nvd.nist.gov/vuln/detail/CVE-2026-54009 advisory
- https://github.com/advisories/GHSA-wch8-mhj5-9frg advisory
- https://github.com/open-webui/open-webui package
- https://github.com/pypa/advisory-database/tree/main/vulns/open-webui/PYSEC-2026-2766.yaml web
- https://pypi.org/project/open-webui web