{
  "id": 6605,
  "url": "https://arxiv.org/abs/2603.28063v1",
  "title": "Reward Hacking as Equilibrium under Finite Evaluation",
  "summary": "We prove that under five minimal axioms -- multi-dimensional quality, finite evaluation, effective optimization, resource finiteness, and combinatorial interaction -- any optimized AI agent will systematically under-invest effort in quality dimensions not covered by its evaluation system. This result establishes reward hacking as a structural equilibrium, not a correctable bug, and holds regardless of the specific alignment method (RLHF, DPO, Constitutional AI, or others) or evaluation architect",
  "authors": "Jiacheng Wang, Jinbin Huang",
  "category": "research",
  "topics": "safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-03-30T06:06:40.000Z",
  "fetched_at": "2026-07-14T16:32:37.308Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/6605",
  "original_url": "https://arxiv.org/abs/2603.28063v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}