CVE-2026-34159: llama.cpp: Unauthenticated RCE via GRAPH_COMPUTE buffer=0 bypass in llama.cpp RPC backend
llama.cpp is an inference of several LLM models in C/C++. Prior to version b8492, the RPC backend's deserializetensor() skips all bounds validation when a tensor's buffer field is 0. An unauthenticated attacker can read and write arbitrary process memory via crafted GRAPHCOMPUTE messages. Combined with pointer leaks from ALLOCBUFFER/BUFFERGETBASE, 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 Software
Remediation
Patch Available
Event History
Frequently Asked Questions
What is the severity of CVE-2026-34159?
CVE-2026-34159 is rated as a critical vulnerability due to the potential for unauthenticated remote code execution.
How do I fix CVE-2026-34159?
To mitigate CVE-2026-34159, upgrade to version b8492 or later of llama.cpp which includes the necessary bounds validation.
What does CVE-2026-34159 affect?
CVE-2026-34159 affects the RPC backend of the llama.cpp software for Large Language Model inference.
Can CVE-2026-34159 be exploited remotely?
Yes, CVE-2026-34159 can be exploited remotely due to the unauthenticated nature of the vulnerability.
What specific component is vulnerable in CVE-2026-34159?
The vulnerable component in CVE-2026-34159 is the deserialize_tensor() function in the llama.cpp RPC backend.