{
  "id": 4673,
  "url": "https://arxiv.org/abs/2605.08827v2",
  "title": "Mental Health AI Safety Claims Must Preserve Temporal Evidence",
  "summary": "The safety of mental health AI is often judged at the wrong temporal scale. Current evaluations typically score isolated responses, endpoint outcomes, or aggregate dialogue quality, while clinically consequential failures may arise from the order and accumulation of interactions themselves, including delayed escalation, repeated reinforcement, dependency formation, failed repair, and gradual deterioration across turns. This paper argues that this mismatch is not merely a limitation of evaluation",
  "authors": "Srimonti Dutta, Ratna Kandala",
  "category": "research",
  "topics": "safety-alignment,healthcare",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-09T09:27:36.000Z",
  "fetched_at": "2026-07-14T16:31:12.743Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/4673",
  "original_url": "https://arxiv.org/abs/2605.08827v2",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}