{
  "id": 3790,
  "url": "https://arxiv.org/abs/2605.24856v1",
  "title": "The Concept Allocation Zone: Tracking How Concepts Form Across Transformer Depth",
  "summary": "Concept formation in transformer language models is depth-extended, not a single-layer event: concepts emerge gradually across a contiguous region of the residual stream. Mechanistic interpretability methods identify the single layer of peak class separation -- the \"best layer\" -- capturing a snapshot rather than the process itself. We introduce the Concept Allocation Zone (CAZ): the depth interval within which a concept becomes measurably separable, the region allocated to its geometric express",
  "authors": "James Henry",
  "category": "research",
  "topics": "safety-alignment,privacy-surveillance",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-24T04:25:32.000Z",
  "fetched_at": "2026-07-14T16:30:31.920Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/3790",
  "original_url": "https://arxiv.org/abs/2605.24856v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}