Post #4168306

Reposting that arxiv paper on the trust issues with inference providers and inference routers. I think the main thing that gets lost on people is that it's not just the model that can do damaging things to your system, but there are ways for the service operators to inject attacks also. And every extra router hop in the inference chain is an additional set of operators that you need to trust. https://arxiv.org/abs/2604.08407v1 #llm