{
  "id": 3756,
  "url": "https://arxiv.org/abs/2605.26174v1",
  "title": "A Universal Cliff and a Design Fingerprint: Cross-Section Defect Detection Under LLM Orchestration",
  "summary": "Production language-model systems answer a request by partitioning it across an invisible orchestration of worker agents that recompose one integrated report. We ask what this does to a class of defect no single worker can see: a contradiction in the relation between two distant sections of a document. Holding the documents, defects, mechanism, scoring, and seed fixed, we vary only the model -- ten systems across five generations from one developer and five providers from distinct alignment para",
  "authors": "Hiroki Fukui",
  "category": "research",
  "topics": "safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-25T05:09:48.000Z",
  "fetched_at": "2026-07-14T16:30:27.612Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/3756",
  "original_url": "https://arxiv.org/abs/2605.26174v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}