GitHub Advisoryadvisory · PyPI
GHSA-94f4-hr76-p5j6 vLLM: OpenAI auth bypass
CRITICAL
Official page ↗
- Aliases
- CVE-2026-48746
- Ecosystem
- PyPI
- Packages
- vllm
- Severity
- Critical
- CVSS vector
- CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H
- Weaknesses
- CWE-444, CWE-501
- Reviewed
- 2026-06-16
Affected packages
• PyPI vllm: >= 0.3.0, < 0.22.0 → fixed in 0.22.0
Description
Summary
A vulnerability in ASGI web servers and starlette's trust on those web servers enables an authentication bypass of the OpenAI API `AuthenticationMiddleware`, which was discovered during @x41sec's source code audit.
It allows to use the API without providing the configured `VLLM_API_KEY` or `--api-key`.
Details
In https://github.com/vllm-project/vllm/blob/v0.14.0/vllm/entrypoints/openai/api_server.py#L689-L692 the `url_path` is taken from the `URL`, which is reconstructed by _starlette_ based on the request `scope`.
```py
from starlette.datastructures import URL, Headers, MutableHeaders, State
...
url_path = URL(scope=scope).path.removeprefix(root_path)
headers = Headers(scope=scope)
if url_path.startswith("/v1") and not self.verify_token(headers):
response = JSONResponse(content={"error": "Unauthorized"}, status_code=401)
return response(scope, receive, send)
return self.app(scope, receive, send)
```
The request `scope` includes the request's `Host:` header and reconstructs the URL as shown below:
```py
f"{scheme}://{host_header}{path}"
```
Neither starlette nor any of the ASGI servers (including uvicorn, which vllm uses) properly filter the `Host:` header for invalid characters. This allows an attacker to include special URL characters such as `/` or `?` in the `Host:` header and thereby control the reconstructed URL and it's `.path` attribute.
FastAPI/starlette's routing uses the HTTP path and does not depend on the parsed url.path attribute, allowing attackers to reach an endpoint via a certain path while providing a different value in the `.path`.
Impact
• Instances of vllm that use an API Key for the OpenAI API and expose the API to attackers.
• Instances behind an RFC-conforming web server (such as nginx) are not affected.
Related
Alias · CVE (1)
- CVE-2026-48746 vLLM is an inference and serving engine for large language models (LLMs). vulnerability
References
- https://github.com/vllm-project/vllm/security/advisories/GHSA-94f4-hr76-p5j6 web
- https://nvd.nist.gov/vuln/detail/CVE-2026-48746 advisory
- https://github.com/vllm-project/vllm/pull/43426 web
- https://x41-dsec.de/lab/advisories/x41-2026-002-starlette web
- https://security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-48746.json web
- https://github.com/vllm-project/vllm package
- https://github.com/pypa/advisory-database/tree/main/vulns/vllm/PYSEC-2026-226.yaml web
- https://bugzilla.redhat.com/show_bug.cgi?id=2491581 web
- https://access.redhat.com/security/cve/CVE-2026-48746 web
- https://access.redhat.com/errata/RHSA-2026:61629 web
- https://access.redhat.com/errata/RHSA-2026:61627 web
- https://access.redhat.com/errata/RHSA-2026:43038 web
- https://access.redhat.com/errata/RHSA-2026:42644 web
- https://access.redhat.com/errata/RHSA-2026:42142 web
- https://access.redhat.com/errata/RHSA-2026:42132 web
- https://access.redhat.com/errata/RHSA-2026:36006 web
- https://access.redhat.com/errata/RHSA-2026:36005 web
- https://access.redhat.com/errata/RHSA-2026:30089 web
- https://access.redhat.com/errata/RHSA-2026:30088 web