CVE-2026-44223

vLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parameters

Published May 12, 2026 CVSS 3.1 · 6.5 Medium EPSS < 1% KEV · Not listed

Description

vLLM is an inference and serving engine for large language models (LLMs). From 0.18.0 to before 0.20.0, the extract_hidden_states speculative decoding proposer in vLLM returns a tensor with an incorrect shape after the first decode step, causing a RuntimeError that crashes the EngineCore process. The crash is triggered when any request in the batch uses sampling penalty parameters (repetition_penalty, frequency_penalty, or presence_penalty). A single request with a penalty parameter (e.g., "repetition_penalty": 1.1) is sufficient to crash the server. This vulnerability is fixed in 0.20.0.

Affected Products

Default status is the baseline for the product, each version can override it (e.g. patched versions marked unaffected).

Vendor Product Default status Versions

vllm-project

vllm

affected

Version	Status	Constraints
`>= 0.18.0, < 0.20.0`	affected	—

Configuration 1 [-]

cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*

No data.

Vendor Product Versions

Vllm-project

Vllm

Version	Status	Scheme	Platform
`[0.18.0,0.20.0)`	affected	semver	—

Remediation

No solution or workaround provided in the CVE record

Tracking

Advisories

Source	ID	Title
Github GHSA	GHSA-83vm-p52w-f9pw	vLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parameters

Metrics

No CVSS v4.0

Attack Vector Network

Attack Complexity Low

Privileges Required Low

Scope Unchanged

Confidentiality Impact None

Integrity Impact None

Availability Impact High

User Interaction None

No CVSS v3.0

No CVSS v2

This CVE is not in the KEV list.

Exploit prediction score: < 1% (0.367% probability within 30 days)

Exploitation poc

Automatable no

Technical Impact partial

References

Link	Providers
https://github.com/vllm-project/vllm/pull/38610
https://github.com/vllm-project/vllm/security/advisories/GHSA-83vm-p52w-f9pw

History

Jun 22, 2026 10:00 PM Description updated View changes

Type	Values Removed	Values Added
Description	vLLM is an inference and serving engine for large language models (LLMs). From to before 0.20.0, the extract_hidden_states speculative decoding proposer in vLLM returns a tensor with an incorrect shape after the first decode step, causing a RuntimeError that crashes the EngineCore process. The crash is triggered when any request in the batch uses sampling penalty parameters (repetition_penalty, frequency_penalty, or presence_penalty). A single request with a penalty parameter (e.g., "repetition_penalty": 1.1) is sufficient to crash the server. This vulnerability is fixed in 0.20.0.	vLLM is an inference and serving engine for large language models (LLMs). From 0.18.0 to before 0.20.0, the extract_hidden_states speculative decoding proposer in vLLM returns a tensor with an incorrect shape after the first decode step, causing a RuntimeError that crashes the EngineCore process. The crash is triggered when any request in the batch uses sampling penalty parameters (repetition_penalty, frequency_penalty, or presence_penalty). A single request with a penalty parameter (e.g., "repetition_penalty": 1.1) is sufficient to crash the server. This vulnerability is fixed in 0.20.0.

May 15, 2026 3:15 PM Scoring metrics updated View changes

Type	Values Removed	Values Added
Metrics		ssvc `{'options': {'Automatable': 'no', 'Exploitation': 'poc', 'Technical Impact': 'partial'}, 'version': '2.0.3'}`

May 14, 2026 3:45 PM First appearance in the feed. Affected versions updated. Affected vendors updated View changes

Type	Values Removed	Values Added
First Time appeared		Vllm Vllm vllm
CPEs		cpe:2.3:a:vllm:vllm::::::::
Vendors & Products		Vllm Vllm vllm

May 12, 2026 11:30 PM First appearance in the feed. Affected vendors updated View changes

Type	Values Removed	Values Added
First Time appeared		Vllm-project Vllm-project vllm
Vendors & Products		Vllm-project Vllm-project vllm

May 12, 2026 8:15 PM Description updated. Title updated. Weaknesses updated. Added references. Scoring metrics updated View changes

Type	Values Removed	Values Added
Description		vLLM is an inference and serving engine for large language models (LLMs). From to before 0.20.0, the extract_hidden_states speculative decoding proposer in vLLM returns a tensor with an incorrect shape after the first decode step, causing a RuntimeError that crashes the EngineCore process. The crash is triggered when any request in the batch uses sampling penalty parameters (repetition_penalty, frequency_penalty, or presence_penalty). A single request with a penalty parameter (e.g., "repetition_penalty": 1.1) is sufficient to crash the server. This vulnerability is fixed in 0.20.0.
Title		vLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parameters
Weaknesses		CWE-131 CWE-704
References		https://github.com/vllm-project/vllm/pull/38610 https://github.com/vllm-project/vllm/security/advisories/GHSA-83vm-p52w-f9pw
Metrics		cvssV3_1 `{'score': 6.5, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H'}`