{
  "id": 5485,
  "url": "https://arxiv.org/abs/2604.20995v2",
  "title": "Value-Conflict Diagnostics Reveal Widespread Alignment Faking in Language Models",
  "summary": "Alignment faking, where a model behaves aligned with developer policy when monitored but reverts to its own preferences when unobserved, is a concerning yet poorly understood phenomenon, in part because current diagnostic tools remain limited. Prior diagnostics rely on highly toxic and clearly harmful scenarios, causing most models to refuse immediately. As a result, models never deliberate over developer policy, monitoring conditions, or the consequences of non-compliance, making these diagnost",
  "authors": "Inderjeet Nair, Jie Ruan, Lu Wang",
  "category": "research",
  "topics": "regulation,safety-alignment,healthcare",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-22T18:37:25.000Z",
  "fetched_at": "2026-07-14T16:31:48.873Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/5485",
  "original_url": "https://arxiv.org/abs/2604.20995v2",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}