CVE-2026-94622

[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, 10:17:00 PM UTC

Source

[NVD] · [CVE record]

EPSS

0.63% · Oct 25, 2026

A flaw was found in vLLM. A remote attacker can exploit this vulnerability by sending requests with incomplete `kv_transfer_params` dictionary entries to the NIXL connector's metadata handling. This can trigger an uncaught error in the EngineCore scheduling, leading to the termination of the decode engine. Consequently, all routed requests will fail until a manual restart, resulting in a denial of service. vllm: vLLM: Denial of Service via Incomplete NIXL KV Transfer Metadata Red Hat rates this vulnerability as Important because an unauthenticated remote attacker can terminate the inference engine process without elevated privileges, halting service for all connected clients until the daemon is manually restarted. In Red Hat OpenShift AI and Red Hat Enterprise Linux AI, this flaw affects distributed deployments running disaggregated prefill and decode roles with the NIXL KV transfer connector enabled. Standalone instances and non-disaggregated vLLM configurations that do not leverage distributed KV cache transfer are not susceptible to this failure. 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.