{
  "id": 6612,
  "url": "https://arxiv.org/abs/2603.27868v1",
  "title": "A Revealed Preference Framework for AI Alignment",
  "summary": "Human decision makers increasingly delegate choices to AI agents, raising a natural question: does the AI implement the human principal's preferences or pursue its own? To study this question using revealed preference techniques, I introduce the Luce Alignment Model, where the AI's choices are a mixture of two Luce rules, one reflecting the human's preferences and the other the AI's. I show that the AI's alignment (similarity of human and AI preferences) can be generically identified in two sett",
  "authors": "Elchin Suleymanov",
  "category": "research",
  "topics": "safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-03-29T20:54:01.000Z",
  "fetched_at": "2026-07-14T16:32:37.309Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/6612",
  "original_url": "https://arxiv.org/abs/2603.27868v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}