CVE-2026-44223
Medium · 6.5CVSS
6.5
EPSS
0.4%
Published
May 12, 2026
CISA due date
—
The record
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 packages
| Ecosystem | Package | Introduced | Fixed |
|---|
Related vulnerabilities
- CVE-2026-106574 ImageMagick is free and open-source software used for editing and manipulating digital… CWE-131
- CVE-2026-88830 A unit confusion in BusyBox TLS Montgomery reduction buffer allocation causes a… CWE-131
- CVE-2026-84448 libheif is a HEIF and AVIF file format decoder and encoder. CWE-131
- CVE-2026-22590 eprosima Fast DDS is a C++ implementation of the DDS (Data Distribution Service) standard… CWE-131
- CVE-2026-78221 An incorrect buffer size calculation in the Windows Interactive Service in OpenVPN… CWE-131
- CVE-2026-18743 A flaw was found in popt. CWE-131
- CVE-2026-78002 A flaw was found in rsyslog. CWE-131
- CVE-2026-16924 IBM AIX 7.2, and 7.3 and IBM PowerVM VIOS 4.1 could allow a remote attacker to cause a… CWE-131