
Cloud Vulnerability DB
A community-led vulnerabilities database
A critical security vulnerability (CVE-2025-32444) has been discovered in vLLM, a high-throughput and memory-efficient inference and serving engine for Large Language Models (LLMs). The vulnerability affects versions from 0.6.5 to 0.8.5 specifically in deployments using the Mooncake integration. This vulnerability has received the highest possible CVSS score of 10.0, indicating its critical severity. The issue was disclosed on April 29, 2025, and has been patched in version 0.8.5 (NVD, Security Online).
The vulnerability stems from the use of pickle-based serialization over unsecured ZeroMQ sockets in vLLM's Mooncake integration. The specific issue is located in the recv_pyobj() function within the vllm/distributed/kv_transfer/kv_pipe/mooncake_pipe.py file, which implicitly uses pickle.loads() to process incoming data over the ZeroMQ sockets. The vulnerable sockets were configured to listen on all network interfaces, significantly increasing the attack surface. The vulnerability has been assigned CWE-502 (Deserialization of Untrusted Data) and received a CVSS v3.1 score of 10.0 with the vector string CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H (GitHub Advisory).
The vulnerability enables remote code execution (RCE) capabilities, potentially allowing attackers to execute arbitrary code on affected systems. Given vLLM's widespread adoption with over 46,000 stars on GitHub and its use across academic, research, and enterprise-grade AI systems, the potential impact is significant. The vulnerability affects all vLLM instances that actively utilize the Mooncake integration, though deployments not using Mooncake are not susceptible (Security Online).
The primary mitigation is to upgrade to vLLM version 0.8.5, which contains the patch for this vulnerability. Organizations using vLLM with Mooncake integration should prioritize this upgrade immediately. For deployments that cannot immediately upgrade, it's worth noting that vLLM instances not using the Mooncake integration are not vulnerable to this specific issue (NVD).
Source: This report was generated using AI
Free Vulnerability Assessment
Evaluate your cloud security practices across 9 security domains to benchmark your risk level and identify gaps in your defenses.
Get a personalized demo
"Best User Experience I have ever seen, provides full visibility to cloud workloads."
"Wiz provides a single pane of glass to see what is going on in our cloud environments."
"We know that if Wiz identifies something as critical, it actually is."