Remote Code Execution Vulnerability in vLLM Integration with Mooncake
CVE-2025-32444
9.8CRITICAL
What is CVE-2025-32444?
vLLM, a memory-efficient inference engine for large language models, has a vulnerability linked to its integration with Mooncake. This issue arises from using insecure ZeroMQ sockets that are open to all network interfaces, making it possible for attackers to execute arbitrary code remotely via pickle-based serialization. Only instances utilizing the Mooncake integration are impacted. The vulnerability has been fully addressed in version 0.8.5.
Affected Version(s)
vllm >= 0.6.5, < 0.8.5