Get real-time updates on Telegram
P4
CVE-2025-71439: llama.cpp 0 to before b5649
CVE-2025-71439. CVSS 3.1 base score 5.5 (ghsa). Fixed version: c33fe8b8c4427202706b1434e9fc8ab5752c9cac.
Affected technology
llama.cpp · 0 to before b5649
ggml-org
Description’s affected range: before b5649
Component: Not specified by the source
Attack conditions (Vendor/CNA, CVSS 4.0): Local · No privileges required · Passive user interaction
What an attacker can do
Attackers can supply a crafted GGUF model with a short tokenizer.ggml.tokens array, causing llama-cli or llama-server to read past id_to_token and crash. Vendor/CNA’s CVSS 4.0 assessment (base score 6.8/10) rates confidentiality and integrity impact as none; availability impact as high.
- CWE
- CWE-125
- CCR priority
- 22.0 /100 (P4)
- CVSS 3.1
- 5.5 /10 · CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:N/I:N/A:H · Vendor/CNA
- EPSS
- 0.00121 · percentile 0.01972 · 2026-11-15
- KEV
- no
Provenance
- FIRST EPSS daily exploit-probability · Source record · observed 2026-11-16 06:41:02.068004+00:00 UTC
- GitHub Advisory Database · Source record · observed 2026-10-27 10:51:29.462336+00:00 UTC
- NVD CVE API 2.0 · Source record · observed 2026-11-05 17:21:24.340896+00:00 UTC
- OSV.dev · Source record · observed 2026-10-27 11:47:01.753480+00:00 UTC
Stable permalink: https://cybercodered.org/item/cve-cve-2025-71439.html