Prompt injection isn’t just a security issue. It reveals how easily AI behavior can be redirected when constraints aren’t enforced.