CVE-2026-52131

A flaw was found in llama.cpp. This vulnerability, triggered by a reachable assertion, could allow an attacker to cause a denial of service (DoS). A successful exploit would make the affected system or application unavailable to legitimate users.

A flaw was found in llama.cpp within the GGUF model reader component (gguf_reader::read). Processing a malformed or specially crafted GGUF model file can trigger a reachable assertion failure during metadata or tensor header parsing. An attacker capable of tricking a user or application into loading a malicious GGUF file can cause the application to terminate unexpectedly, resulting in a denial of service.

The CVSS score(s) listed for this vulnerability do not reflect the associated product's status, and are included for informational purposes to better understand the severity of this vulnerability.

Restrict model loading permissions to prevent unprivileged users from providing custom or untrusted GGUF model files to llama.cpp processes. Validate and ingest GGUF model files only from trusted, authenticated registry or storage sources.