An uncontrolled memory allocation in the GGUF metadata parser allows a crafted GGUF file to trigger an unrecoverable Go makeslice panic, causing the entire Ollama server process to terminate. The vulnerability is triggered by length and count fields in string lengths, tensor dimensions, and metadata arrays that are used as allocation sizes without validation against the remaining file size, resulting in a fatal crash without recovery.

The affected product is Ollama, the open‑source model hosting platform. Any deployment of Ollama that accepts GGUF uploads through the blob upload, model create, or pull API endpoints is vulnerable. The impact is not limited to running the HEAD f0078ae codebase will be affected until the parser is corrected.

The CVSS score of 8.7 classifies this as high very low but non‑zero likelihood of exploitation in the wild, and the vulnerability is not listed in the CISA KEV catalog. Attackers can target the server remotely by submitting a sub‑1KB GGUF file via the documented API endpoints, making this a purely remote denial of service threat that bypasses existing recovery middleware and forces a complete process shutdown.