Skip to content
Submit incident
AII-1001-91977
ConfirmedA court, a regulator or the organization itself has stated the core facts, or the record cites a court or regulator source.

vLLM Remote Code Execution via Mooncake Integration (CVE-2025-32444)

vLLM contains a critical remote code execution vulnerability in its Mooncake integration due to unsafe pickle deserialization over exposed ZeroMQ sockets. The vulnerability affects versions 0.6.5 and later and was patched in version 0.8.5.

19 March 2025 Β· vLLM Project Β· Security and cyberattacks
19 March 2025
AII-1001-91977View source β†—
Quick facts
Category
Security and cyberattacks
Source
github.com
Date
19 March 2025
Drafted by
Raidu Incident Detection

What happened

vLLM versions 0.6.5 and later are vulnerable to remote code execution via the Mooncake integration. The vulnerability exists in vllm/distributed/kv_transfer/kv_pipe/mooncake_pipe.py where recv_pyobj() contains implicit pickle.loads(), which can deserialize untrusted data. The vulnerable ZeroMQ sockets listen on all network interfaces without authentication controls, allowing attackers to send malicious payloads over the network.

Reported impact

The vulnerability allows remote code execution on distributed hosts that use Mooncake to distribute KV cache across distributed environments. Any vLLM deployments using Mooncake to distribute KV across distributed hosts are impacted by this vulnerability.

Index analysis

The issue is similar to a prior Mooncake integration vulnerability (GHSA-x3m8-f7g5-qhm7) also involving unsafe pickle deserialization over exposed ZeroMQ sockets. vLLM instances that do not make use of the Mooncake integration are not vulnerable.

Governance controls

The vulnerability was patched in vLLM v0.8.5, addressing affected versions 0.6.5 and later.

Classification

Category
Security and cyberattacks
Industry
IT infrastructure and hosting
Harm status
Potential harm
Intent
Unintentional
Timing
Unknown
AI systems
vLLM

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

github.com
https://github.com/vllm-project/vllm/security/advisories/GHSA-hj4w-hm2g-p6w5