JFrog disclosed CVE-2026-105192 on 7 October 2026, a critical flaw in LMCache, the open-source key-value cache that speeds up LLM serving engines such as vLLM. In multiprocess (distributed) mode LMCache runs as a standalone server and opens a ZeroMQ socket, port 5555 by default, so worker processes can register and share cached blocks. That socket has no authentication. One message type carries a msgpack extension that is handed to Python's pickle.loads while the server is still decoding the REGISTER_KV_CACHE arguments, before any handler or type check runs, so a single crafted message executes the sender's code as the LMCache user.

JFrog says the official container images run that process as root. The 9.8 score applies when the server is bound to a routable address with --host, which is the documented multi-node setup and what LMCache's own example Kubernetes DaemonSet does (it listens on every interface); a default single-host install that stays on localhost is not reachable from other machines, and LMCache running inside a single vLLM process does not open the port at all.

The decode path arrived in 0.3.9 (October 2025) and is still present in 0.5.5, the latest PyPI release, the 0.5.6 release candidates through rc3 and the dev branch as of 7 October.

LMCache has not published its own advisory, and JFrog gives no way to tell whether a server has already been hit. Separately, one GitHub account opened six more LMCache security reports on 6 October alleging unauthenticated cross-tenant cache access and command-running network services; those have no CVE or maintainer confirmation. Researcher: Yuval Moravchick (JFrog).

Primary: JFrog Security Research advisory JFSA-2026-001694382; wire: The Hacker News (7 Oct).

Correction (22 Oct 2026): this card said no fixed release existed and that LMCache had not published its own advisory. LMCache published advisory GHSA-w5f2-9q9c-ff4h on 13 October (CVSS 3.1 9.8) listing 0.3.9 up to 0.5.6 as affected and 0.5.6 as the fix; 0.5.6 was released on PyPI on 15 October. (source: LMCache GitHub advisory GHSA-w5f2-9q9c-ff4h)