Securing AI Agents Against Prompt Injection Attacks