Every CVE Search/ Get API key

CVE-2026-48746

vLLM: OpenAI auth bypass

9.1 CriticalFix available

vLLM is an inference and serving engine for large language models (LLMs). From 0.3.0 until 0.22.0, a vulnerability in ASGI web servers and starlette's trust on those web servers enables an authentication bypass of the OpenAI API AuthenticationMiddleware. It allows to use the API without providing the configured VLLM_API_KEY or --api-key. This vulnerability is fixed in 0.22.0.

Attack profile

How easy it is to exploit

Attack vector
Network
Complexity
Low
Privileges needed
None
User interaction
None

What an attacker can affect

Confidentiality
High
Integrity
None
Availability
High

Exploit likelihood

1.15%Up 1.1 points since Jun 12, 2026
90%50%10%1%

What each source reports

CVE List Score used 9.1

vLLM: OpenAI auth bypass

vLLM is an inference and serving engine for large language models (LLMs). From 0.3.0 until 0.22.0, a vulnerability in ASGI web servers and starlette's trust on those web servers enables an authentication bypass of the OpenAI API AuthenticationMiddleware. It allows to use the API without providing the configured VLLM_API_KEY or --api-key. This vulnerability is fixed in 0.22.0.

Record
CVE-2026-48746
CVSS 3.1
9.1 CriticalCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H
Weakness
CWE-444, CWE-501
Updated
2026-11-02
EUVDEUVD-2026-38401 9.1

vLLM is an inference and serving engine for large language models (LLMs). From 0.3.0 until 0.22.0, a vulnerability in ASGI web servers and starlette's trust on those web servers enables an authentication bypass of the OpenAI API AuthenticationMiddleware. It allows to use the API without providing the configured VLLM_API_KEY or --api-key. This vulnerability is fixed in 0.22.0.

Record
EUVD-2026-38401
CVSS 3.1
9.1 CriticalCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H
Updated
2026-11-02
GitHub AdvisoriesGHSA-94f4-hr76-p5j6 9.1

vLLM: OpenAI auth bypass

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.

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:

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.
Record
GHSA-94f4-hr76-p5j6
CVSS 3.1
9.1 CriticalCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H
Weakness
CWE-444, CWE-501
Updated
2026-11-02
NVD 9.1

vLLM is an inference and serving engine for large language models (LLMs). From 0.3.0 until 0.22.0, a vulnerability in ASGI web servers and starlette's trust on those web servers enables an authentication bypass of the OpenAI API AuthenticationMiddleware. It allows to use the API without providing the configured VLLM_API_KEY or --api-key. This vulnerability is fixed in 0.22.0.

Record
CVE-2026-48746
CVSS 3.1
9.1 CriticalCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H
Weakness
CWE-444, CWE-501
Updated
2026-11-02
OSV 9.1

vLLM: OpenAI auth bypass

vLLM is an inference and serving engine for large language models (LLMs). From 0.3.0 until 0.22.0, a vulnerability in ASGI web servers and starlette's trust on those web servers enables an authentication bypass of the OpenAI API AuthenticationMiddleware. It allows to use the API without providing the configured VLLM_API_KEY or --api-key. This vulnerability is fixed in 0.22.0.

Record
CVE-2026-48746
CVSS 3.1
9.1 CriticalCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H
Weakness
CWE-444
Updated
2026-11-02
OSVGHSA-94f4-hr76-p5j6 9.1

vLLM: OpenAI auth bypass

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.

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:

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.
Record
GHSA-94f4-hr76-p5j6
CVSS 3.1
9.1 CriticalCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H
Weakness
CWE-444, CWE-501
Updated
2026-11-02
OSVPYSEC-2026-226 9.1

vLLM is an inference and serving engine for large language models (LLMs). From 0.3.0 until 0.22.0, a vulnerability in ASGI web servers and starlette's trust on those web servers enables an authentication bypass of the OpenAI API AuthenticationMiddleware. It allows to use the API without providing the configured VLLM_API_KEY or --api-key. This vulnerability is fixed in 0.22.0.

Record
PYSEC-2026-226
CVSS 3.1
9.1 CriticalCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H
Updated
2026-11-02

Affected products

Show 4 moreShow less

Affected packages

Affected packages
PackageAffectedFixed
vllm
PyPI
0.3.0 up to 0.22.0
0.22.0
vllm
PyPI
0.3.0 up to 0.22.0, plus listed versions
0.10.00.10.10.10.1.10.10.20.11.0
Show 45 moreShow less
0.11.10.11.20.12.00.13.00.14.00.14.10.15.00.15.10.16.00.17.00.17.10.18.00.18.10.19.00.19.10.20.00.20.10.20.20.21.00.3.10.3.20.3.30.4.00.4.0.post10.4.10.4.20.4.30.5.00.5.0.post10.5.10.5.20.5.30.5.3.post10.5.40.5.50.6.00.6.10.6.1.post10.6.1.post20.6.20.6.30.6.3.post10.6.40.6.4.post10.6.5
0.22.0

History

History
RecordedChangedSources
first indexedCVE List, EUVD, GitHub Advisories, NVD, OSV

Dates are when Every CVE recorded each change. Upstream dates may differ.

In short

CVE-2026-48746 is a critical vulnerability (CVSS 9.1) in Vllm, published June 16, 2026. No source lists public exploit code. EPSS puts the chance of exploitation in the next 30 days at 1.15%, the 67th percentile. Fixed in vllm 0.22.0.

References

Show 15 moreShow less
This page doesn't list everything

Left out here: 17 versions. The JSON record has them.

MoveEnter Open