Hacking Back the AI-Hacker: Prompt Injection as a Defense for LLM-Attackers