Nice paper about why LLMs can never be fully protected against prompt injection. https://role-confusion.github.io/ #LLM #AI #Promptinjection #cybersecurity