by Kastro
Paper/Archive/CVE-2026-34159

CVE-2026-34159

ggml-org llama.cpp — llama.cpp: Unauthenticated RCE via GRAPH_COMPUTE buffer=0 bypass in llama.cpp RPC backend

EPSS 2%public exploitCVSS 9.8
critical.is priorityof 100
55
CVSS 9.8 (CNA)9.8 / 10 × 3029.4Not in CISA KEV+300.0EPSS percentile 76.1.761 × 2015.2Exploit: SSVC poc+1010.0SSVC exploitation: poc+100.0
Description · from the CVE record

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.

Affected

ProductAffectedFixed inSource
llama.cpp< b8492–CNA

Scores by source

GitHub_M (CNA)9.8CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
CISA ADPSSVCExploitation: poc · Automatable: yes · Technical impact: total

References

refgithub.com/ggml-org/llama.cpp/security/advisories/GHSA-j8rj-fmpv-wcxwrefgithub.com/ggml-org/llama.cpp/pull/20908refgithub.com/ggml-org/llama.cpp/commit/39bf0d3c6a95803e0f41aaba069ffbee26721042