{
  "id": 5300,
  "url": "https://arxiv.org/abs/2604.24902v1",
  "title": "Safety Drift After Fine-Tuning: Evidence from High-Stakes Domains",
  "summary": "Foundation models are routinely fine-tuned for use in particular domains, yet safety assessments are typically conducted only on base models, implicitly assuming that safety properties persist through downstream adaptation. We test this assumption by analyzing the safety behavior of 100 models, including widely deployed fine-tunes in the medical and legal domains as well as controlled adaptations of open foundation models alongside their bases. Across general-purpose and domain-specific safety b",
  "authors": "Emaan Bilal Khan, Amy Winecoff, Miranda Bogen, Dylan Hadfield-Menell",
  "category": "research",
  "topics": "healthcare",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-27T18:34:08.000Z",
  "fetched_at": "2026-07-14T16:31:40.219Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/5300",
  "original_url": "https://arxiv.org/abs/2604.24902v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}