vLLM - Denial of Service via Unvalidated Multimodal Embeddings
Summary
| CVE | CVE-2026-56340 |
|---|---|
| State | PUBLISHED |
| Assigner | VulnCheck |
| Source Priority | CVE Program / NVD first with legacy fallback |
| Published | 2026-06-20 19:16:23 UTC |
| Updated | 2026-06-30 03:21:04 UTC |
| Description | vLLM versions >= 0.10.2 and < 0.13.0 are missing sparse tensor validation in multimodal embeddings processing. Because PyTorch disables sparse tensor invariant checks by default, an attacker can submit crafted embedding requests with malformed (negative or out-of-bounds) tensor indices, when the prompt-embeds feature is enabled, to trigger crashes or resource exhaustion (denial of service), with potential for out-of-bounds/write-what-where memory corruption. This continues CVE-2025-62164, whose prior fix only disabled the feature by default rather than addressing the root cause. |
Risk And Classification
Primary CVSS: v4.0 8.7 HIGH from [email protected]
CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/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
EPSS: 0.003520000 probability, percentile 0.272080000 (date 2026-07-04)
Problem Types: CWE-20 | NVD-CWE-noinfo | CWE-787 | CWE-20 Improper Input Validation | CWE-787 Out-of-bounds Write
| Version | Source | Type | Score | Severity | Vector |
|---|---|---|---|---|---|
| 4.0 | [email protected] | Secondary | 8.7 | HIGH | CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N/E:X/C... |
| 4.0 | CNA | CVSS | 8.7 | HIGH | CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N |
| 3.1 | [email protected] | Primary | 7.5 | HIGH | CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H |
| 3.1 | ADP | CVSS | 8.8 | HIGH | CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H |
| 3.1 | [email protected] | Secondary | 8.8 | HIGH | CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H |
| 3.1 | 0b0ca135-0b70-47e7-9f44-1890c2a1c46c | Secondary | 8.8 | HIGH | CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H |
| 3.1 | CNA | CVSS | 8.8 | HIGH | CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H |
CVSS v4.0 Breakdown
CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/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
CVSS v3.1 Breakdown
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
NVD Known Affected Configurations (CPE 2.3)
Vendor Declared Affected Products
| Source | Vendor | Product | Version | Platforms |
|---|---|---|---|---|
| CNA | VLLM | VLLM | affected 0.10.2 0.13.0 semver | Not specified |
| CNA | VLLM | VLLM | unaffected 0.13.0 semver | Not specified |
| ADP | Red Hat | Red Hat AI Inference Server | Not specified | Not specified |
| ADP | Red Hat | Red Hat Enterprise Linux AI RHEL AI 3 | Not specified | Not specified |
| ADP | Red Hat | Red Hat OpenShift AI RHOAI | Not specified | Not specified |
References
| Reference | Source | Link | Tags |
|---|---|---|---|
| www.vulncheck.com/advisories/vllm-denial-of-service-via-unvalidated-multimodal-... | [email protected] | www.vulncheck.com | Third Party Advisory |
| access.redhat.com/security/cve/CVE-2026-56340 | 0b0ca135-0b70-47e7-9f44-1890c2a1c46c | access.redhat.com | |
| security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-56340.json | 0b0ca135-0b70-47e7-9f44-1890c2a1c46c | security.access.redhat.com | |
| github.com/vllm-project/vllm/security/advisories/GHSA-mcmc-2m55-j8jj | [email protected] | github.com | Vendor Advisory |
| bugzilla.redhat.com/show_bug.cgi | 0b0ca135-0b70-47e7-9f44-1890c2a1c46c | bugzilla.redhat.com | |
| CVE Program record | CVE.ORG | www.cve.org | canonical |
| NVD vulnerability detail | NVD | nvd.nist.gov | canonical, analysis |
Additional Advisory Data
| Source | Time | Event |
|---|---|---|
| ADP | 2026-06-20T19:01:01.538Z | Reported to Red Hat. |
| ADP | 2026-06-20T18:27:10.148Z | Made public. |
Workarounds
ADP: Disable prompt-embeds if not required. Restrict who can submit multimodal embedding requests. Apply authentication and rate limits on inference APIs. Upgrade to a fixed vLLM build when available from Red Hat.