CVE-2026-34159
Description
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.
Predictions
Heuristic predictions, AS-IS, for prioritization only.
Mitigations
Vendor advisory: debian — https://security-tracker.debian.org/tracker/CVE-2026-34159
Vendor advisory: security-advisories@github.com — https://github.com/ggml-org/llama.cpp/security/advisories/GHSA-j8rj-fmpv-wcxw
Vendor advisory: security-advisories@github.com — https://github.com/ggml-org/llama.cpp/pull/20908
Vendor advisory: security-advisories@github.com — https://github.com/ggml-org/llama.cpp/commit/39bf0d3c6a95803e0f41aaba069ffbee26721042
OS impact
| OS | Version | Status | Fixed in |
|---|---|---|---|
| debian | forky | fixed | 8611+dfsg-1 |
| debian | sid | fixed | 8611+dfsg-1 |
Application impact
| Vendor | Product | Versions | Fixed |
|---|---|---|---|
| ggml | llama.cpp | {"endExcluding":"b8492"} | b8492 |
References
CWEs
CWE-119
Verify integrity in audit chain (admin only). AS-IS.