{
  "id": 6105,
  "url": "https://arxiv.org/abs/2604.08524v1",
  "title": "What Drives Representation Steering? A Mechanistic Case Study on Steering Refusal",
  "summary": "Applying steering vectors to large language models (LLMs) is an efficient and effective model alignment technique, but we lack an interpretable explanation for how it works-- specifically, what internal mechanisms steering vectors affect and how this results in different model outputs. To investigate the causal mechanisms underlying the effectiveness of steering vectors, we conduct a comprehensive case study on refusal. We propose a multi-token activation patching framework and discover that dif",
  "authors": "Stephen Cheng, Sarah Wiegreffe, Dinesh Manocha",
  "category": "research",
  "topics": "safety-alignment,finance-investment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-09T17:57:14.000Z",
  "fetched_at": "2026-07-14T16:32:15.636Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/6105",
  "original_url": "https://arxiv.org/abs/2604.08524v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}