Path-Aware Defense Against Prompt Injection Attacks in Retrieval-Augmented Large Language Models. JAAIR 2026, 1 (1).