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

MITRE NVD CVE.ORG JSON API Print: PDF PDF

Summary

CVECVE-2026-44223
StatePUBLISHED
AssignerGitHub_M
Source PriorityCVE Program / NVD first with legacy fallback
Published2026-05-12 20:16:43 UTC
Updated2026-05-12 20:16:43 UTC
DescriptionvLLM 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.

Risk And Classification

Primary CVSS: v3.1 6.5 MEDIUM from [email protected]

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H

Problem Types: CWE-131 | CWE-704 | CWE-131 CWE-131: Incorrect Calculation of Buffer Size | CWE-704 CWE-704: Incorrect Type Conversion or Cast

VersionSourceTypeScoreSeverityVector
3.1[email protected]Secondary6.5MEDIUMCVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H
3.1CNADECLARED6.5MEDIUMCVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H

CVSS v3.1 Breakdown

Attack Vector
Network
Attack Complexity
Low
Privileges Required
Low
User Interaction
None
Scope
Unchanged
Confidentiality
None
Integrity
None
Availability
High

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H

Vendor Declared Affected Products

SourceVendorProductVersionPlatforms
CNA Vllm-project Vllm affected >= 0.18.0, < 0.20.0 Not specified

References

ReferenceSourceLinkTags
github.com/vllm-project/vllm/pull/38610 [email protected] github.com
github.com/vllm-project/vllm/security/advisories/GHSA-83vm-p52w-f9pw [email protected] github.com
CVE Program record CVE.ORG www.cve.org canonical
NVD vulnerability detail NVD nvd.nist.gov canonical, analysis
© CVE.report 2026 |

Use of this information constitutes acceptance for use in an AS IS condition. There are NO warranties, implied or otherwise, with regard to this information or its use. Any use of this information is at the user's risk. It is the responsibility of user to evaluate the accuracy, completeness or usefulness of any information, opinion, advice or other content. EACH USER WILL BE SOLELY RESPONSIBLE FOR ANY consequences of his or her direct or indirect use of this web site. ALL WARRANTIES OF ANY KIND ARE EXPRESSLY DISCLAIMED. This site will NOT BE LIABLE FOR ANY DIRECT, INDIRECT or any other kind of loss.

CVE, CWE, and OVAL are registred trademarks of The MITRE Corporation and the authoritative source of CVE content is MITRE's CVE web site. This site includes MITRE data granted under the following license.

Free CVE JSON API cve.report/api

CVE.report and Source URL Uptime Status status.cve.report