We are generally aware that the RPC backend andrpc-server are vulnerable to all sorts of attack vectors. At some point we added a notice in the README to avoid usage of the RPC functionality in sensitive environments:https://github.com/ggml-org/llama.cpp/tree/master/examples/rpc

However, we keep receiving security advisories about the RPC backend that we don't have the capacity to act upon privately. It makes more sense for the time being to resolve such vulnerabilities publicly, so that the community can help in the process.

With this change to the security policy, we categorize such issues as known vulnerabilities and recommend to skip the advisory process. My suggestion is this to be in effect until we feel more confident about the security of the RPC implementation.