{
  "id": 448,
  "url": "https://arxiv.org/abs/2606.30219v1",
  "title": "EvalSafetyGap: A Hybrid Survey and Conceptual Framework for LLM Evaluation-Safety Failures",
  "summary": "LLM evaluation and AI safety face a shared measurement problem: benchmark scores, reward-model signals, and reported safety metrics can improve while the latent properties they are meant to represent remain difficult to verify. This paper combines a hybrid survey - a systematic search paired with narrative synthesis and separately tracked grey evidence - with a conceptual framework and a structured ten-model audit. The synthesis spans eight evidence streams: benchmark validity, dynamic evaluatio",
  "authors": "Buğra Alperen Uluırmak, Rifat Kurban",
  "category": "research",
  "topics": "safety-alignment,transparency",
  "orgs": null,
  "regions": null,
  "published_at": "2026-06-29T12:33:06.000Z",
  "fetched_at": "2026-07-14T14:14:32.647Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/448",
  "original_url": "https://arxiv.org/abs/2606.30219v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}