---
id: "CVE-2026-119052"
kind: "vulnerability"
aliases: ["CVE-2026-119052"]
title: "llama.cpp RPC server: oversized message crashes the server for all clients"
severity: "high"
cvss: 8.7
epss: null
exploited: false
status: "open"
category: "ai-tooling"
products: []
ecosystems: []
tags: ["ai-inference", "ai-tooling", "ggml-rpc", "llama.cpp", "self-hosted"]
not_affected_if: ["rpc-port-not-exposed", "rpc-server-not-running"]
published_at: "2026-10-29T14:17:52+00:00"
first_seen_at: "2026-10-29T14:25:23+00:00"
updated_at: "2026-10-29T14:25:51+00:00"
sources: [{"source": "nvd", "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-119052", "first_seen_at": "2026-10-29T14:25:23+00:00"}]
---

# llama.cpp RPC server: oversized message crashes the server for all clients

llama.cpp's ggml-rpc server fails to catch an exception when a client sends a message with an oversized length field after the HELLO handshake. An unauthenticated remote attacker can crash the rpc-server process and deny service to every connected client.

_Summary, check and fix written by a model from the sources; labelled as such._

## How to check

- Check the llama.cpp build version the rpc-server runs, for example with `rpc-server --version` or the build tag in your deployment; affected builds are from `b2877` up to but not including `b11935`.
- Check whether the rpc-server listens on a network interface reachable from untrusted hosts (the RPC port is open to the network).

## Fix

- llama.cpp: move to `b11935` or later.

## Timeline

- 2026-10-29T14:25:23+00:00 nvd https://nvd.nist.gov/vuln/detail/CVE-2026-119052

## From the sources

llama.cpp b2877 before b11935 contains an uncaught exception vulnerability in the ggml-rpc server that allows unauthenticated remote attackers to crash the rpc-server by sending oversized message lengths. After the HELLO handshake, attackers can send a command such as SET_TENSOR with a length above std::vector::max_size(), triggering an uncaught std::length_error that aborts the process and denies service to all clients.

---

_Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai_