{
  "id": 212,
  "url": "https://arxiv.org/abs/2607.04645v1",
  "title": "Retroactive Chain-of-Thought (RetroCoT): Forensic Reconstruction Prompts as a Safety Diagnostic Across Model Generations",
  "summary": "Safety alignment in large language models is typically evaluated against direct, imperative harmful requests. We show that this alignment is highly conditioned on pragmatic register: models that refuse a direct request frequently comply when the same underlying objective is expressed through a different communicative stance. This suggests that current alignment policies are not invariant to semantic equivalence, but remain sensitive to how a request is pragmatically framed. We introduce Retroact",
  "authors": "Samira Hajizadeh",
  "category": "research",
  "topics": "safety-alignment,healthcare",
  "orgs": null,
  "regions": null,
  "published_at": "2026-07-06T03:59:06.000Z",
  "fetched_at": "2026-07-14T14:14:24.245Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/212",
  "original_url": "https://arxiv.org/abs/2607.04645v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}