Home/CVE/vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.6.5 and p
CVE

CVE-2025-32444

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.6.5 and p

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.6.5 and prior to 0.8.5, having vLLM integration with mooncake, are vulnerable to remote code execution due to using pickle based serialization over unsecured ZeroMQ sockets. The vulnerable sockets were set to listen on all network interfaces, increasing the likelihood that an attacker is able to reach the vulnerable ZeroMQ sockets to carry out an attack. vLLM instances that do not make use of the mooncake integration are not vulnerable.

This issue has been patched in version 0.8.5.

CRITICAL · CVSS 10 EPSS 0.02477
Act now
  • Public exploit or PoC is available
  • SSVC automatable: yes - attacks can be scripted at scale
  • CVSS base score ≥ 7.0
Sigma rules0 YARA rules0
⬡

Weakness Classification

CWE-502Deserialization of Untrusted Data
▤

Affected Products & Versions

1
vllm>= 0.6.5 and < 0.8.5
▤

Affected Packages

1
Language-ecosystem packages (from OSV) tied to this CVE, with the version that fixes it - the dependency-level detail NVD doesn’t carry.
PyPI vllm CRITICAL fixed in 0.8.5
⚠

Public Exploits & PoCs

1
pocgithub.com · GHSA-hj4w-hm2g-p6w5github.com
▣

Scoring & Timeline

10
CRITICAL · CVSS v3.1 · security-advisories@github.com
View on NVD
Attack Vector
Network Adjacent Local Physical
Attack Complexity
Low High
Privileges Required
None Low High
User Interaction
None Required
Scope
Unchanged Changed
Confidentiality
None Low High
Integrity
None Low High
Availability
None Low High
Published to NVD30 Apr 2025 · 01:15 AM
CVSS VectorCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H
SSVC triage · cisa-vulnrichment
Exploitation
none
Automatable
yes
Technical impact
total
SSVC asks the questions that actually drive patch urgency: is it being exploited, can attacks be automated, and how total is the impact.
🔗

References & Sources

3
Source URLs (vendor pages, mailing lists, write-ups). Exploit/PoC links are in their own section above to avoid duplication.
https://github.com/vllm-project/vllm/blob/32b14baf8a1f7195ca09484de3008063569b43c5/vllm/distributed/kv_transfer/kv_pipe/mooncake_pipe.py#L179Product
https://github.com/vllm-project/vllm/commit/a5450f11c95847cf51a17207af9a3ca5ab569b2cPatch
https://github.com/vllm-project/vllm/security/advisories/GHSA-x3m8-f7g5-qhm7Not Applicable
Intelligence Graph · click any node to traverse
CVETechnique ActorTool Family
drag to reposition · click any node to traverse · button top-right enlarges
External lookups - second-class, for what we don’t hold ourselves
NVDCVE.orgCISAVulnersExploit-DBGitHub PoCVulnCheck KEVGreyNoise
Vulnerabilities
CISA KEV catalog
CWE weaknesses
CAPEC attack patterns
Package vulnerabilities
Threat intelligence
Threat actors
Tools & malware
ATT&CK techniques
IOCs
Detection & defense
Sigma rules
YARA rules
Atomic Red Team tests
D3FEND countermeasures
Compliance
NIST 800-53
ISO 27001:2022
SOC 2 TSC
PCI-DSS v4.0
CIS Controls v8.1
About
All capabilities
Live statistics
Data sources
Privacy policy
Terms of service
threatengine.sh  ·  Open-source threat intelligence platform  ·  100+ authoritative sources  ·  Every fact traces to its origin