{
  "id": 3814,
  "url": "https://arxiv.org/abs/2605.24614v1",
  "title": "Measuring the Depth of LLM Unlearning via Activation Patching",
  "summary": "Large language model (LLM) unlearning has emerged as a crucial post-hoc mechanism for privacy protection and AI safety, yet auditing whether target knowledge is truly erased remains challenging. Existing output-level metrics fail to detect when this knowledge remains recoverable from internal representations. Recent white-box studies reveal such residual knowledge but often rely on auxiliary training or dataset-specific adaptations, leaving no generalizable metric. To address these limitations, ",
  "authors": "Jaeung Lee, Dohyun Kim, Jaemin Jo",
  "category": "research",
  "topics": "safety-alignment,privacy-surveillance,transparency",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-23T14:52:28.000Z",
  "fetched_at": "2026-07-14T16:30:31.921Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/3814",
  "original_url": "https://arxiv.org/abs/2605.24614v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}