Design Patterns for Securing LLM Agents against Prompt Injections

This new paper by 11 authors from organizations including IBM, Invariant Labs, ETH Zurich, Google and Microsoft is an excellent addition to the literature on prompt injection and LLM security. …