Unpatched Critical LMCache Flaw Lets Unauthenticated Attackers Run Code Remotely

A critical vulnerability in LMCache, open-source software that speeds up large language model (LLM) servers such as vLLM, lets an attacker run code on the cache server without logging in, and no fixed version is available.

The flaw is in LMCache's multiprocess mode, where the cache runs as a standalone server that LLM workers reach over the ZeroMQ messaging library. A single network connection is sufficient for an unauthenticated attacker to execute arbitrary code on the cache server. The vulnerability has been reported in October 2026, and no patched version or fixed release is currently available.

LMCache is widely integrated with LLM serving frameworks including vLLM. Successful exploitation could allow an attacker to gain control of the cache server, potentially leading to full compromise of the LLM serving infrastructure, data exfiltration, or disruption of AI services. Because the flaw allows remote code execution without authentication, the risk is considered significant for any organization running LMCache in multiprocess mode with network access.