CVE-2026-53923

Severity CVSS v4.0:

MEDIUM

Type:

CWE-200 Information Leak / Disclosure

Publication date:

22/06/2026

Last modified:

24/06/2026

Description

vLLM is an inference and serving engine for large language models (LLMs). From 0.5.5 until 0.23.1rc0, integer truncation of tensor dimensions in vLLM&#39;s GGUF dequantize kernels (csrc/quantization/gguf/gguf_kernel.cu) causes partial tensor processing. The output tensor is allocated at full size via torch::empty (uninitialized memory), but the dequantize CUDA kernel processes only a truncated number of elements. The unfilled portion of the output tensor retains whatever was previously in GPU memory. In multi-tenant inference deployments, this residual GPU memory may contain tensor data from other users&#39; inference requests, constituting information disclosure. This vulnerability is fixed in 0.23.1rc0.

Impact

Vector 4.0

CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:L/VI:L/VA:N/SC:N/SI:N/SA:N CVSS v4.0 Severity and Metrics:

Base Score: 5.30 MEDIUM
Vector: CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:L/VI:L/VA:N/SC:N/SI:N/SA:N

Attack Vector (AV): Network
Attack Complexity (AC): Low
Attack Requirements (AT): None
Privileges Required (PR): None
User Interaction (UI): Passsive
Confidentiality (VC): Low
Integrity (VI): Low
Availability (VA): None
Confidentiality (SC): None
Integrity (SI): None
Availability (SA): None

Base Score 4.0

5.30

Severity 4.0

MEDIUM

Vector 3.x

CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N CVSS v3.1 Severity and Metrics:

Base Score: 7.50 HIGH
Vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N

Attack Vector (AV): Network
Attack Complexity (AC): Low
Privileges Required (PR): None
User Interaction (UI): None
Scope (S): Unchanged
Confidentiality (C): High
Integrity (I): None
Availability (A): None