- CVSS
- HIGH · 8.1v3.1CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H
- Published
- 2026-10-28
- Weakness
- CWE-787
- Source
- nvd.nist.gov/vuln/detail/CVE-2026-118375
Description
llama.cpp b8841 before b11865 contains an out-of-bounds write vulnerability in the ggml-rpc RDMA transport rdma_recv() function that allows unauthenticated peers to overflow the rpc-server stack. Attackers advertising RDMA capabilities in the HELLO message can send up to 256 KiB into a 1-byte stack variable in rpc_serve_client(), crashing the server or potentially executing code.
References
- https://github.com/ggml-org/llama.cpp
- https://github.com/ggml-org/llama.cpp/blob/620d2225cc436ab0b001e3200e590e3b286ef9ab/ggml/src/ggml-rpc/transport.cpp#L510-L528
- https://github.com/ggml-org/llama.cpp/commit/9a0fde9967fa1a35b6e5dd1f47e1a3648b6fecc1
- https://github.com/ggml-org/llama.cpp/pull/30399
- https://github.com/ggml-org/llama.cpp/releases/tag/b11865
- https://www.vulncheck.com/advisories/llama-cpp-b8841-before-b11865-stack-based-buffer-overflow-via-ggml-rpc-rdma-rdma-recv
How GTK Cyber trains on this
AI security training at GTK Cyber covers the LLM and ML-pipeline vulnerability classes that vulnerabilities like CVE-2026-118375 fall into. Our hands-on courses are taught by Charles Givre and other practitioners who break and defend production AI systems.