CVE-2026-94626

[CVE catalog]

Severity

High · CVSS 8.7CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X

Published

Sep 21, 2026, 10:00:00 PM UTC

Updated

Sep 21, 2026, 10:17:01 PM UTC

Source

[NVD] · [CVE record]

EPSS

0.63% · Oct 25, 2026

A flaw was found in vLLM. A remote attacker can exploit a vulnerability in the `kv_transfer_params` function on OpenAI-compatible completion endpoints by supplying an unvalidated `tp_size` parameter. This allows for unbounded memory allocation, leading to memory exhaustion and a kernel Out-Of-Memory (OOM) kill of the decode worker process. This ultimately results in a Denial of Service (DoS) for the affected system. vllm: vLLM: Denial of Service via unvalidated memory allocation parameter This vulnerability is rated as Important severity because unauthenticated remote attackers can cause an out-of-memory denial of service condition without special privileges, terminating active inference worker processes. In Red Hat OpenShift AI and Red Hat AI Inference Server deployments utilizing disaggregated prefill and decode architectures, completion endpoints exposed to untrusted networks can receive crafted transfer parameters that trigger excessive memory consumption. This memory exhaustion causes the host kernel to abort the decode worker, interrupting model serving for all users on the affected node. The CVSS score(s) listed for this vulnerability do not reflect the associated product's status, and are included for informational purposes to better understand the severity of this vulnerability.