Your llama.cpp Server May Be Leaking Auth Tokens — This Patch Fixes It

The llama.cpp project, known for enabling local LLM inference on consumer hardware, has tagged version b9738. This release addresses a critical security issue in the server component: **auth headers were being forwarded in CORS proxy requests**, potentially exposing sensitive authentication tokens to unintended third-party origins. The fix, co-authored by Xuan Son Nguyen from Hugging Face, ensures that when llama.cpp acts as a proxy, it no longer blindly forwards authorization headers—a common vector for credential theft.

The b9738 release ships with the usual broad platform support: macOS (Apple Silicon and Intel), Linux (x64/arm64 with CPU, Vulkan, ROCm, OpenVINO, SYCL), Windows (x64/arm64 with CUDA 12 & 13, Vulkan, OpenVINO, SYCL, HIP), and Android arm64. Notably, Apple Silicon builds include KleidiAI optimizations, while openEuler builds remain disabled. This release is a patch-only update with no new features, but the security fix is critical for anyone running llama.cpp's server in a production or multi-tenant environment where CORS proxying is enabled.