{
  "id": 5362,
  "url": "https://arxiv.org/abs/2604.23445v1",
  "title": "AI Safety Training Can be Clinically Harmful",
  "summary": "Large language models are being deployed as mental health support agents at scale, yet only 16% of LLM-based chatbot interventions have undergone rigorous clinical efficacy testing, and simulations reveal psychological deterioration in over one-third of cases. We evaluate four generative models on 250 Prolonged Exposure (PE) therapy scenarios and 146 CBT cognitive restructuring exercises (plus 29 severity-escalated variants), scored by a three-judge LLM panel. All models scored near-perfectly on",
  "authors": "Suhas BN, Andrew M. Sherrill, Rosa I. Arriaga, Chris W. Wiese, Saeed Abdullah",
  "category": "research",
  "topics": "safety-alignment,healthcare,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-25T21:09:50.000Z",
  "fetched_at": "2026-07-14T16:31:40.223Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/5362",
  "original_url": "https://arxiv.org/abs/2604.23445v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}