I like the way this "Design Patterns for Securing LLM Agents against Prompt Injections" paper puts it: https://
simonwillison.net/2025/Jun/13/pr
ompt-injection-design-patterns/#scope-of-the-problem
…

→ View original post on X — @simonw