CVE-2026-118375: llama.cpp out-of-bounds write in ggml-rpc RDMA transport

CVE-2026-118375 · Severity: high · CVSS 8.1 · Published 2026-10-28

llama.cpp is a C/C++ library for running large language model inference. An unauthenticated remote attacker can send a specially crafted RDMA message to a running RPC server, causing a stack buffer overflow that crashes the service or potentially executes arbitrary code on the host.

Technologies: Ggml-Org Llama.Cpp. Vendors: Ggml-Org.

Quick answers

Is CVE-2026-118375 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-118375 being exploited in the wild.
How severe is CVE-2026-118375?
High severity, with a CVSS score of 8.1, published 28 October 2026.
Which products and versions are affected?
Ggml-Org Llama.Cpp before b11865.
Is there a fix for CVE-2026-118375?
Yes: fixed in b11865. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-118375 attacked?
It is attacked over the network, in a software library.
Am I affected by CVE-2026-118375?
Check whether you run Ggml-Org Llama.Cpp at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

How to fix

  • Fixed in: b11865

Technical details

The ggml-rpc RDMA transport's rdma_recv() function fails to validate buffer sizes when processing HELLO messages with advertised RDMA capabilities, allowing an attacker to write up to 256 KiB into a 1-byte stack variable in rpc_serve_client(). This out-of-bounds write occurs without authentication, leveraging the network-accessible RPC server interface.

Affected products

  • ggml-org llama.cpp before b11865

Timeline

  • 2026-10-28: disclosed

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.