How do you handle possible prompt injection in emails?