Ai Security

Prompt injection is a boundary problem, not just a wording problem

Applications that read external text need a clear distinction between information to inspect and instructions they are allowed to follow.