CVE-2026-21869
High · 8.8CVSS
8.8
EPSS
0.5%
Published
Jan 8, 2026
CISA due date
—
The record
llama.cpp is an inference of several LLM models in C/C++. In commits 55d4206c8 and prior, the n_discard parameter is parsed directly from JSON input in the llama.cpp server's completion endpoints without validation to ensure it's non-negative. When a negative value is supplied and the context fills up, llama_memory_seq_rm/add receives a reversed range and negative offset, causing out-of-bounds memory writes in the token evaluation loop. This deterministic memory corruption can crash the process or enable remote code execution (RCE). There is no fix at the time of publication.
Affected packages
| Ecosystem | Package | Introduced | Fixed |
|---|---|---|---|
| openSUSE:Tumbleweed | llamacpp | 0 | 9500-1.1 |
Related vulnerabilities
- CVE-2026-111497 A buffer overflow vulnerability in the Brotli decoder (c/dec/decode.c) in Google Brotli… CWE-787
- CVE-2026-112397 Pillow is a Python imaging library. CWE-787
- CVE-2026-112059 libLAS through 1.8.1 contains a heap out-of-bounds write vulnerability in the point… CWE-787
- CVE-2026-89088 A flaw was found in parted. CWE-787
- CVE-2026-89085 A flaw was found in parted. CWE-787
- CVE-2026-85746 MariaDB server is a community developed fork of MySQL server. CWE-787
- CVE-2026-107815 MariaDB server is a community developed fork of MySQL server. CWE-787
- CVE-2026-56851 The Nickname profile can panic with an out-of-bounds slice error when transforming… CWE-787