Skip to main content


This is a masterful demonstration of the unsecurable nature of LLMs, and how prompt injection by dedicated humans who know how to write will always win.

role-confusion.github.io/