What you didn’t want to know about prompt injections in LLM apps
blog.ml6.eu
I have developed attacks that can bypass all existing defences.