Latest vulnerabilities for "vllm vllm"

vllm vllmvLLM denial of service via prompt embeds on M-RoPE models

Risk 40

Severity

7.1

First published (updated )

CVE-2026-55514

vllm vllmvLLM: ReDoS via structured_outputs.regex compiled without timeout in xgrammar and outlines backends

Risk 47

Severity

8.7

First published (updated )

CVE-2026-55574

vllmvLLM: Remote DoS in vLLM via Invalid Recovered Token Reinjection

Risk 43

Severity

7.5

First published (updated )

CVE-2026-54234

vllm vllmvLLM speech-to-text endpoints allocate full upload before enforcing the audio file-size limit

Risk 38

Severity

6.5

First published (updated )

CVE-2026-55646

vllm vllmvLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.0, an assert…

Risk 33

Severity

7

First published (updated )

REDHAT-BUG-2491582

pypi/vllmvLLM: Dependency Confusion Vulnerability in vLLM Dockerfile

Risk 77

Severity

8.8

First published (updated )

CVE-2026-54232

pypi/vllmvLLM - Denial of Service via Unvalidated Multimodal Embeddings

Risk 79

Severity

8.7

First published (updated )

CVE-2026-56340

pypi/vllmvllm - Regular Expression Denial of Service in Multiple Components

Risk 43

Severity

5.3

First published (updated )

CVE-2025-71379

pip/vllmvLLM: OOM Denial of Service via Audio Decompression Bomb

Risk 38

Severity

6.5

First published (updated )

CVE-2026-54233

pip/vllmvLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via Anthropic router

Risk 27

Severity

5.3

First published (updated )

CVE-2026-54236

pip/vllmvLLM GGUF Kernels: int64_t to int truncation of tensor dimensions causes GPU buffer overflow

Risk 43

Severity

5.3

First published (updated )

CVE-2026-53923

pip/vllmvLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernels

Risk 33

Severity

6.9

First published (updated )

CVE-2026-54235

pip/vllmvLLM: OpenAI auth bypass

Risk 66

Severity

9.1

First published (updated )

CVE-2026-48746

pip/vllmvLLM: Security Check Bypass via assert Statement in Activation Function Loading Allows Arbitrary Code Execution

Risk 69

Severity

7.5

First published (updated )

CVE-2026-41523

vllm vllmvLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory (OOM) Denial of Service (DoS) attac…

Risk 33

Severity

7

First published (updated )

REDHAT-BUG-2487813

pypi/vllmUnbounded Frame Count in video/jpeg Base64 Data URL Processing Leads to OOM DoS in vllm-project/vllm

Risk 43

Severity

7.5

First published (updated )

CVE-2026-5497

pip/vllmvLLM: Artifact Pin Decay in vLLM allows pinned deployments to load unpinned code, weights, and processors

Risk 45

Severity

6.5

First published (updated )

CVE-2026-47155

pip/vllmvLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parameters

Risk 38

Severity

6.5

First published (updated )

CVE-2026-44223

pip/vllmvLLM: Remote DoS via Special-Token Placeholders

Risk 43

Severity

7.5

First published (updated )

CVE-2026-44222

pypi/vllmvllm KV Block kv_cache_interface.py has_mamba_layers uninitialized resource

Risk 31

Severity

2.9

EPSS

0.07%

First published (updated )

CVE-2026-7141

pip/vllmvLLM Affected by Denial of Service via Unbounded Frame Count in video/jpeg Base64 Processing

Risk 38

Severity

6.5

First published (updated )

CVE-2026-34755

pip/vllmvLLM affected by Server-Side Request Forgery (SSRF) in `download_bytes_from_url `

Risk 34

Severity

5.4

First published (updated )

CVE-2026-34753

pip/vllmvLLM Affected by Unauthenticated OOM Denial of Service via Unbounded `n` Parameter in OpenAI API Server

Risk 38

Severity

6.5

First published (updated )

CVE-2026-34756

pypi/librosavLLM: Downmix Implementation Differences as Attack Vectors Against Audio AI Models

Risk 48

Severity

7.1

First published (updated )

CVE-2026-34760

pypi/vllmvLLM's hardcoded trust_remote_code=True in NemotronVL and KimiK25 bypasses user security opt-out

Risk 77

Severity

8.8

First published (updated )

CVE-2026-27893

pip/vllmSSRF Protection Bypass in vLLM

Risk 86

Severity

9.8

First published (updated )

CVE-2026-25960

pip/vllmvLLM leaks a heap address when PIL throws an error

Risk 61

Severity

9.8

EPSS

0.05%

First published (updated )

CVE-2026-22778

vllmvLLM vulnerable to Server-Side Request Forgery (SSRF) in `MediaConnector`

Risk 35

Severity

7.1

EPSS

0.01%

First published (updated )

CVE-2026-24779

pip/vllmvLLM affected by RCE via auto_map dynamic module loading during model initialization

Risk 61

Severity

9.8

EPSS

0.05%

First published (updated )

CVE-2026-22807

vllmvLLM is vulnerable to DoS in Idefics3 vision models via image payload with ambiguous dimensions

Risk 31

Severity

7.5

EPSS

0.05%

First published (updated )

CVE-2026-22773

Trending

Versions

vllm vllmvLLM denial of service via prompt embeds on M-RoPE models

vllm vllmvLLM: ReDoS via structured_outputs.regex compiled without timeout in xgrammar and outlines backends

vllmvLLM: Remote DoS in vLLM via Invalid Recovered Token Reinjection

vllm vllmvLLM speech-to-text endpoints allocate full upload before enforcing the audio file-size limit

vllm vllmvLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.0, an assert…

Don't miss critical vulnerabilities

pypi/vllmvLLM: Dependency Confusion Vulnerability in vLLM Dockerfile

pypi/vllmvLLM - Denial of Service via Unvalidated Multimodal Embeddings

pypi/vllmvllm - Regular Expression Denial of Service in Multiple Components

pip/vllmvLLM: OOM Denial of Service via Audio Decompression Bomb

pip/vllmvLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via Anthropic router

Don't miss critical vulnerabilities

pip/vllmvLLM GGUF Kernels: int64_t to int truncation of tensor dimensions causes GPU buffer overflow

pip/vllmvLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernels

pip/vllmvLLM: OpenAI auth bypass

pip/vllmvLLM: Security Check Bypass via assert Statement in Activation Function Loading Allows Arbitrary Code Execution

vllm vllmvLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory (OOM) Denial of Service (DoS) attac…

Don't miss critical vulnerabilities

pypi/vllmUnbounded Frame Count in video/jpeg Base64 Data URL Processing Leads to OOM DoS in vllm-project/vllm

pip/vllmvLLM: Artifact Pin Decay in vLLM allows pinned deployments to load unpinned code, weights, and processors

pip/vllmvLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parameters

pip/vllmvLLM: Remote DoS via Special-Token Placeholders

pypi/vllmvllm KV Block kv_cache_interface.py has_mamba_layers uninitialized resource

Don't miss critical vulnerabilities

pip/vllmvLLM Affected by Denial of Service via Unbounded Frame Count in video/jpeg Base64 Processing

pip/vllmvLLM affected by Server-Side Request Forgery (SSRF) in `download_bytes_from_url `

pip/vllmvLLM Affected by Unauthenticated OOM Denial of Service via Unbounded `n` Parameter in OpenAI API Server

pypi/librosavLLM: Downmix Implementation Differences as Attack Vectors Against Audio AI Models

pypi/vllmvLLM's hardcoded trust_remote_code=True in NemotronVL and KimiK25 bypasses user security opt-out

Don't miss critical vulnerabilities

pip/vllmSSRF Protection Bypass in vLLM

pip/vllmvLLM leaks a heap address when PIL throws an error

vllmvLLM vulnerable to Server-Side Request Forgery (SSRF) in `MediaConnector`

pip/vllmvLLM affected by RCE via auto_map dynamic module loading during model initialization

vllmvLLM is vulnerable to DoS in Idefics3 vision models via image payload with ambiguous dimensions

Don't miss critical vulnerabilities

Company

Resources

Contact