{
  "id": 6220,
  "url": "https://arxiv.org/abs/2605.26438v1",
  "title": "LURE: Live-Usage Replay Evaluations for Reducing Evaluation Awareness",
  "summary": "Large language models can recognize when they are being evaluated (evaluation awareness) and behave differently because of that, which undermines the validity of safety and alignment benchmarks. We propose LURE (Live-Usage Replay Evaluations), a method for constructing deployment-like evaluations by replaying realistic agentic interaction trajectories and appending evaluation prompt at the end. We also introduce an automated pipeline for measuring evaluation realism, combining detection of verba",
  "authors": "Igor Ivanov, David Demitri Africa",
  "category": "research",
  "topics": "safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-08T00:04:19.000Z",
  "fetched_at": "2026-07-14T16:32:20.055Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/6220",
  "original_url": "https://arxiv.org/abs/2605.26438v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}