The issue originates from a reachable assertion located in the gguf_reader::read function of llama.cpp. Because the assertion is reachable, an attacker could provide crafted input that triggers the assertion, causing the program to terminate unexpectedly. This leads to a denial‑of‑service condition where the affected service or application stops responding. The type of weakness is an assertion misuse that fails to perform proper error handling, which can destabilise the entire process.

The vulnerability affects the llama.cpp library for all versions at or before commit b5693. Operators of systems that rely on this library for loading GGUF model files are impacted. The specific product is the open‑source llama.cpp repository, managed by ggml-org. No vendor‑specific product names are supplied, but any deployment of the library that imports GGUF models and has not applied newer versions is at risk.

The CVSS score is 7.5, and the EPSS score of less than 1% suggests a low exploitation probability. The issue is not listed in the CISA KEV catalog, implying no known large‑scale exploit activity yet. Nonetheless, a reachable assertion that can trigger a crash can be exploited by sending malicious GGUF files over a network or local filesystem if the application does not protect against them. The susceptibility is high if the application runs with elevated privileges or manages critical resources.