PT-2026-50144

Name of the Vulnerable Software and Affected Versions vLLM versions 0.3.0 through 0.21.0 Description An authentication bypass exists in the OpenAI API AuthenticationMiddleware due to improper trust in the reconstructed URL path from the ASGI scope. The url path is derived from a URL object...

GHSA-WCWG-C5FC-9VRC vLLM is vulnerable to an Out-of-Memory (OOM) Denial of Service (DoS) attack due to unbounded frame count processing in the `VideoMediaIO.load_base64()` method

vLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory OOM Denial of Service DoS attack due to unbounded frame count processing in the VideoMediaIO.loadbase64 method. When processing video/jpeg data URLs, the method splits the base64 data string on commas to extract individual JPEG fram...

vLLM is vulnerable to an Out-of-Memory (OOM) Denial of Service (DoS) attack due to unbounded frame count processing in the `VideoMediaIO.load_base64()` method

vLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory OOM Denial of Service DoS attack due to unbounded frame count processing in the VideoMediaIO.loadbase64 method. When processing video/jpeg data URLs, the method splits the base64 data string on commas to extract individual JPEG fram...

CVE-2026-5497

vLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory OOM Denial of Service DoS attack due to unbounded frame count processing in the VideoMediaIO.loadbase64 method. When processing video/jpeg data URLs, the method splits the base64 data string on commas to extract individual JPEG fram...

Allocation of Resources Without Limits or Throttling

Overview vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs Affected versions of this package are vulnerable to Allocation of Resources Without Limits or Throttling via the VideoMediaIO.loadbase64 function. An attacker can cause the server to exhaust memory...

CVE-2026-5497

vLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory OOM Denial of Service DoS attack due to unbounded frame count processing in the "VideoMediaIO.loadbase64" method. When processing "video/jpeg" data URLs, the method splits the base64 data string on commas to extract individual JPEG...

PT-2026-80324

vLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory OOM Denial of Service DoS attack due to unbounded frame count processing in the VideoMediaIO.load base64 method. When processing video/jpeg data URLs, the method splits the base64 data string on commas to extract individual JPEG...

CVE-2026-5497: Uncontrolled Resource Consumption

vLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory OOM Denial of Service DoS attack due to unbounded frame count processing in the VideoMediaIO.loadbase64 method. When processing video/jpeg data URLs, the method splits the base64 data string on commas to extract individual JPEG fram...

Use of Incorrectly-Resolved Name or Reference

Overview vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs Affected versions of this package are vulnerable to Use of Incorrectly-Resolved Name or Reference through several model loading paths. An attacker can make the server load a different Hugging Face...

CVE-2026-4944

A flaw was found in vllm-project/vllm. A hardcoded setting in the model implementation files bypasses the user's explicit security configuration, which is intended to prevent the execution of remote code. This allows a remote attacker to achieve remote code execution by providing a malicious mode...

CVE-2026-9540

A flaw was found in vllm-project vllm, specifically within its OpenAI-compatible Serving Path. A remote attacker could exploit this vulnerability by manipulating certain processing, leading to a denial of service DoS. This could make the affected service unavailable to legitimate users. The issue...

external-log-connector (>=0.1.1, <=0.1.2), lmc-external-log-backend (0.1.0) +3 more potentially affected by CVE-2026-10813 via lmcache (>=0.4.3, <=0.5.5)

lmcache PYPI version =0.4.3, =0.1.1, =0.6.1rc1, =1.0.8, =1.0.9 Source cves: CVE-2026-10813 Source advisory: SNYK:PYTHON-LMCACHE-17675615...

CVE-2026-44222 vulnerabilities

Vulnerabilities for packages: vllm-openai-cuda-12.9, tritonserver-backend-vllm-cuda-13.0...

CVE-2026-7141 vulnerabilities

Vulnerabilities for packages: py3-vllm-cuda-12.9, tritonserver-backend-vllm-cuda-13.0...

GHSA-X368-4G9H-FVV4 vulnerabilities

Vulnerabilities for packages: py3-vllm-cuda-12.9, tritonserver-backend-vllm-cuda-13.0...

aana (>=0.2.1, <=0.2.2.2), acai-swarm (0.1.0) +214 more potentially affected by CVE-2026-4944 via vllm (>=0.1.3, <=0.30.0)

vllm PYPI version =0.1.3, =0.2.1, =1.2.1, =0.0.0, =2.3.5.dev0, =0.0.7, =0.0.1b1, =0.2.2, =0.2.4, =1.0.0, =1.0.2 and more Source cves: CVE-2026-4944 Source advisory: SNYK:PYTHON-VLLM-17675624...

CVE-2026-4944

vllm-project/vllm version 0.14.1 contains a vulnerability where the trustremotecode=True parameter is hardcoded in two model implementation files vllm/modelexecutor/models/nemotronvl.py and vllm/modelexecutor/models/kimik25.py. This bypasses the user's explicit --trust-remote-code=False setting,...

CVE-2026-4944 Hardcoded trust_remote_code=True in vllm-project/vllm Bypasses User Security Control

vllm-project/vllm version 0.14.1 contains a vulnerability where the trustremotecode=True parameter is hardcoded in two model implementation files vllm/modelexecutor/models/nemotronvl.py and vllm/modelexecutor/models/kimik25.py. This bypasses the user's explicit --trust-remote-code=False setting,...

CVE-2026-4944 Hardcoded trust_remote_code=True in vllm-project/vllm Bypasses User Security Control

vllm-project/vllm version 0.14.1 contains a vulnerability where the trustremotecode=True parameter is hardcoded in two model implementation files vllm/modelexecutor/models/nemotronvl.py and vllm/modelexecutor/models/kimik25.py. This bypasses the user's explicit --trust-remote-code=False setting,...

CVE-2026-4944

The provided documents describe a vulnerability in vllm-project/vllm version 0.14.1 where trust_remote_code is hardcoded to True in nemotron_vl.py and kimi_k25.py, bypassing user-specified --trust-remote-code=False and enabling remote code execution via malicious HuggingFace model repositories. T...