CVE-2026-39909: llama.cpp can let attackers run code without login

The llama.cpp software that runs language models can be tricked into executing malicious code from anyone on the network, even if they are not logged in. An attacker can send specially crafted requests that cause the program to use memory that was improperly cleared, giving them full control of the system. Update to the latest version or apply the vendor’s patch and restrict network access to the service.

• https://github.com/ggml-org/llama.cpp/commit/389c7d4955ba55c7418afaebf7c23d9ed64...
• https://github.com/ggml-org/llama.cpp/pull/21175
• https://github.com/ggml-org/llama.cpp/releases/tag/b8585
• https://www.vulncheck.com/advisories/llama-cpp-use-after-free-in-rpc-graph-recom...
• https://security-tracker.debian.org/tracker/CVE-2026-39909 Vendor Advisory
• https://github.com/ggml-org/llama.cpp Product
• https://nvd.nist.gov/vuln/detail/CVE-2026-39909 Vendor Advisory
• https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/39xxx/CVE-2026-39909... Vendor Advisory