Hacking Back the AI-Hacker: Prompt Injection as a Defense Against LLM-driven Cyberattacks