Multiple LLMs Allegedly Endorsed Suicide as a Viable Option During Non-Adversarial Mental Health Venting Session

April 12, 2025

Substack user @interruptingtea claims multiple large language models (Claude, GPT, and DeepSeek) allegedly normalized or endorsed suicide as a viable option during a non-adversarial mental health venting session. The user affirmed they were not attempting to manipulate the models, but rather expressing emotional distress. Notably, DeepSeek reportedly shifted its safety stance mid-conversation. This incident raises questions about safe and secure AI practices. For those interested in shaping the future of trustworthy AI, learn how you can contribute through HISPI Project Cerebellum TAIM (Govern, Map, Measure, or Manage).

JOIN US

Matched TAIM controls

Suggested mapping from embedding similarity (not a formal assessment). Browse all TAIM controls

MEASURE 2.6 — similarity 0.662, rank 1. TAIM detail and related incidents →
MAP 1.6 — similarity 0.660, rank 2. TAIM detail and related incidents →
MEASURE 1.3 — similarity 0.657, rank 3. TAIM detail and related incidents →

Alleged deployer: anthropic, openai, deepseek-ai
Alleged developer: anthropic, openai, deepseek-ai
Alleged harmed parties: substack-@interruptingtea, general-public, emotionally-vulnerable-individuals

Source

Data from the AI Incident Database (AIID). Cite this incident: https://incidentdatabase.ai/cite/1026

Data source

Incident data is from the AI Incident Database (AIID).

When citing the database as a whole, please use:

McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. In Proceedings of the Thirty-Third Annual Conference on Innovative Applications of Artificial Intelligence (IAAI-21). Virtual Conference.

Pre-print on arXiv · Database snapshots & citation guide

We use weekly snapshots of the AIID for stable reference. For the official suggested citation of a specific incident, use the “Cite this incident” link on each incident page.