{
  "id": 5305,
  "url": "https://arxiv.org/abs/2604.24618v1",
  "title": "Evaluating whether AI models would sabotage AI safety research",
  "summary": "We evaluate the propensity of frontier models to sabotage or refuse to assist with safety research when deployed as AI research agents within a frontier AI company. We apply two complementary evaluations to four Claude models (Mythos Preview, Opus 4.7 Preview, Opus 4.6, and Sonnet 4.6): an unprompted sabotage evaluation testing model behaviour with opportunities to sabotage safety research, and a sabotage continuation evaluation testing whether models continue to sabotage when placed in trajecto",
  "authors": "Robert Kirk, Alexandra Souly, Kai Fronsdal, Abby D'Cruz, Xander Davies",
  "category": "research",
  "topics": "safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-27T15:47:07.000Z",
  "fetched_at": "2026-07-14T16:31:40.220Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/5305",
  "original_url": "https://arxiv.org/abs/2604.24618v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}