AI Stack Alert Medium CVE-2026-126975 For AppSec, AI Eng, SOC / IR AI-assisted
ZDI-26-976: llama.cpp GGUF file parsing heap buffer overflow allows remote code execution (CVE-2026-126975)
A heap-based buffer overflow exists in llama.cpp's parsing of GGUF files. The flaw results from missing validation of user-supplied data length before it is copied to a heap buffer. Exploitation requires the target to visit a malicious page or open a malicious file, and llama.cpp has issued an update to correct it.
Why it matters
Organizations running affected llama.cpp installations could have arbitrary code executed in the context of the current process if a user opens a malicious GGUF file or visits a malicious page. Attackers would gain code execution on hosts running local model inference.
Do this
- Identify llama.cpp deployments and their versions, including embedded or bundled copies in AI applications.
- Update llama.cpp to the fixed version referenced in upstream pull request 31336 or later.
- Restrict loading of GGUF model files to trusted sources and verify file provenance before use.
- Hunt for unexpected GGUF files or unexpected child processes spawned by llama.cpp-based services.
Exploit Clock
No signals yet
- No exploitation signals yet
Signals
- NoCISA KEV
- 26/100Priority
- 70%Confidence
Evidence
“The issue results from the lack of proper validation of the length of user-supplied data prior to copying it to a heap-based buffer.”
“User interaction is required to exploit this vulnerability in that the target must visit a malicious page or open a malicious file.”
“llama.cpp has issued an update to correct this vulnerability.”
Sources
Affected / related
llama.cpp, GGUF file parser
Published 2026-11-13 06:00 UTC · updated 2026-11-13 21:35 UTC · AI-assisted summary, checked against quoted sources · How we score