Vulnerability in vLLM Affecting Large Language Models by Vendor vLLM
CVE-2025-62372

8.3HIGH

Key Information:

Vendor

Vllm-project

Status
Vllm
Vendor
CVE Published:
21 November 2025

What is CVE-2025-62372?

A vulnerability in the vLLM engine allows attackers to crash the inference engine by submitting multimodal embedding inputs with valid dimensions but incorrect shapes. This affects any models, regardless of their specific input configurations. The issue has been resolved in version 0.11.1, making it crucial for users to upgrade to this version to safeguard their systems from potential disruptions.

Affected Version(s)

vllm >= 0.5.5, < 0.11.1

References

https://github.com/vllm-project/vllm/security/advisories/...
x_refsource_CONFIRM
https://github.com/vllm-project/vllm/pull/27204
x_refsource_MISC
https://github.com/vllm-project/vllm/pull/6613
x_refsource_MISC

CVSS V4

Score:
8.3
Severity:
HIGH
Confidentiality:
None
Integrity:
None
Availability:
High
Attack Vector:
Network
Attack Complexity:
Low
Attack Required:
None
Privileges Required:
Undefined
User Interaction:
None

Timeline

  • Vulnerability published

  • Vulnerability Reserved

JSON
.
CVE-2025-62372 : Vulnerability in vLLM Affecting Large Language Models by Vendor vLLM