Shipped: H-Node Attack and Defense in Large Language Models

Identifies hallucination-linked internal nodes in LLMs, shows how they can be exploited, and introduces defenses against those attacks.