Buffer overflow in Ggml Llama.cpp
CVE-2026-34159
llama.cpp is an inference of several LLM models in C/C++. Prior to version b8492, the RPC backend's deserialize_tensor() skips all bounds validation when a tensor's buffer field is 0. An unauthenticated attacker can read and write arbitrary process memory via crafted GRAPH_COMPUTE messages. Combined with pointer leaks from ALLOC_BUFFER/BUFFER_GET_BASE, this gives full ASLR bypass and remote code execution. No authentication required, just TCP access to the RPC server port. This issue has been patched in version b8492.
Vulnerability class: Buffer Overflow
Published · last modified .
CVSS v3 metric
CVSS v3 base score 9.8 (Critical). Vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H.
EPSS exploit prediction
EPSS 0.016, 76.5th percentile (scored ); unchanged over 30 days. About EPSS.
Affected products
- Ggml Llama.cpp
- Ggml-Org Llama.cpp — versions < b8492
Weakness classification (CWE)
Public exploits and proof-of-concept code
3 public repositories with exploit or proof-of-concept code for this CVE:
- casp3r0x0/CVE-2026-34159 (28 stars, last pushed 2026-04-23): 0 Click RCE exploit for CVE-2026-34159 Lama.cpp RPC server
- TeamN4C/SG-2026-0021 (last pushed 2026-08-27): CVE-2026-34159 PoC and exploit
- rohithronanki/CVE-2026-34159-Vulnerability-Research-Analysis-Detection (last pushed 2026-05-04): Critical buffer validation bypass in deserialize_tensor() (llama.cpp < b8492). Null tensor buffer skips bounds check, enabling unauthenticated arbitrary R/W via GRAPH_COMPUTE. ASLR bypass + RCE over…
References
- https://github.com/ggml-org/llama.cpp/security/advisories/GHSA-j8rj-fmpv-wcxw (Exploit, Vendor Advisory)
- https://github.com/ggml-org/llama.cpp/pull/20908 (Patch, Issue Tracking)
- https://github.com/ggml-org/llama.cpp/commit/39bf0d3c6a95803e0f41aaba069ffbee26721042 (Patch)
This CVE has public exploit evidence. See the latest CVEs with public exploits.