I've seen a lot of people ranting about AI and LLM injections, but I'm not seeing a conversation about the very obvious threat to AI itself, which brings a possible new era of attacks into the fold.
That is IPoIB and RoCE, being used and are keeping AI infrastructure running properly on what looks to be a very fragile system. There is a major attack surface here.
What resources can I find on this subject outside the RFC's?