{
  "id": 110,
  "url": "https://arxiv.org/abs/2607.07903v1",
  "title": "Mechanistic Interpretability of LLM Jailbreaks via Internal Attribution Graphs",
  "summary": "Large language models (LLMs) exhibit remarkable capabilities but remain highly vulnerable to adversarial prompts and jailbreak attacks. Existing approaches primarily analyze these failures through input-output behaviors or attribution methods, offering limited insight into how adversarial perturbations alter the model's internal reasoning. Consequently, the mechanisms underlying unsafe or incorrect behaviors remain poorly understood. We introduce a mechanistic framework for diagnosing LLM vulner",
  "authors": "Anupam Wagle, Ifrat Ikhtear Uddin, Chaowei Zhang, Longwei Wang",
  "category": "research",
  "topics": "safety-alignment,healthcare",
  "orgs": null,
  "regions": null,
  "published_at": "2026-07-08T20:31:06.000Z",
  "fetched_at": "2026-07-14T14:14:19.967Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/110",
  "original_url": "https://arxiv.org/abs/2607.07903v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}