Hi everyone! We spent a lot of time tinkering about prompt injection detection tools, i.e. what works and what doesn't and also recently released our own prompt injection scanner which greatly outperforms the previous state of the art.
One of the core things we realized is the challenge of evaluating existing alternatives for practitioners that are in need of a solution like LLM Guard. For that reason, we decided to release a benchmark that allows you to prompt and test a set of prompt injection scanner tools ranging from LLM Guard, Rebuff, to our very own. You can also read more about the solution here:
https://laiyer.substack.com/p/how-do-prompt-injection-scanne...