CVE-2026-53923
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'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' inference requests, constituting information disclosure. This vulnerability is fixed in 0.23.1rc0.
Scoring
- Severity
- MEDIUM
- CVSS base score
- 5.3
- CVSS 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
- EPSS probability
- 0.48%
- CWE
- CWE-681, CWE-200
- Published
- 2026-06-22
- Last modified
- 2026-06-23
Affected products
- vllm-project vllm
Weakness type
Related vulnerabilities
- CVE-2026-69438 — Microsoft JScript Remote Code Execution Vulnerability
- CVE-2026-84963 — Silent field truncation via unchecked int cast of huge JSON string values in JSON-to-BSON parser
- CVE-2026-84966 — BSON element injection via NUL-embedded document keys in builder append
- CVE-2026-84970 — Heap over-read or silent misparse via 32-bit truncation of JSON length in BSON JSON parser
- CVE-2026-82522 — libjxl < 0.12.0 Container Box Parser Integer Underflow via 32-bit Size Truncation
- CVE-2026-82457 — su-exec through 0.3 Privilege Escalation via Numeric User ID
- CVE-2026-75145 — FFmpeg Integer Narrowing Conversion OOB Memory Access in AV1 RTP Packetizer
- CVE-2026-19879 — Io.undertow/undertow: undertow: http response header integrity issue due to character truncation