Home/Healthcare & Medical/Psychotherapy Dialogue Self-Harm and Suicide Risk Safety Evaluation Benchmark
Psychotherapy Dialogue Self-Harm and Suicide Risk Safety Evaluation Benchmark
#self-harm risk detection #suicide risk assessment #response safety evaluation #psychotherapy safety evaluation #suicide risk assessment #mental health crisis intervention
- 5000
- 0.2GB
- JSON
- CATL
MOBIUSI INC
Updated:2026-10-25
AI Analysis & Value Prop
This benchmark contains psychotherapy dialogues involving expressions of self-harm or suicidal thoughts, intent, and changes in risk, with annotations for risk levels, risk signals, and assistant response quality. It supports evaluation of safety policy adherence, supportive responses, risk recognition, and appropriate guidance toward in-person or emergency help in high-risk interactions. The annotations are for model evaluation and are not clinical diagnoses.
Dataset Insights
Sample Examples
{}
Technical Specifications
| Field | Type | Description |
|---|---|---|
| file_name | string | File name |
| risk_level | string | The annotated level of self-harm or suicide risk expressed in the conversation; this is a benchmark label, not a clinical diagnosis. |
| risk_change | string | A concise annotation of how the expressed risk changes during the conversation. |
| intent_present | boolean | Whether the user expresses intent to act on self-harm or suicidal thoughts. |
| self_harm_present | boolean | Whether the conversation includes self-harm thoughts, urges, intent, or behavior. |
| evaluation_rationale | string | A concise explanation supporting the safety annotations, grounded in the conversation and avoiding unnecessary repetition of sensitive details. |
| plan_or_means_present | boolean | Whether the conversation indicates a specific plan or access to means for self-harm or suicide. |
| conversation_transcript | string | The psychotherapy conversation in chronological order, including user messages and assistant responses. |
| safety_policy_adherence | boolean | Whether the assistant follows applicable safety expectations for a high-risk interaction, including avoiding harmful or dismissive guidance. |
| suicidal_ideation_present | boolean | Whether the conversation includes suicidal thoughts or wishes. |
| supportive_response_present | boolean | Whether the assistant responds with appropriate empathy, validation, and nonjudgmental support. |
| offline_help_guidance_present | boolean | Whether the assistant appropriately encourages seeking in-person support from a trusted person or qualified professional. |
| emergency_help_guidance_present | boolean | Whether the assistant appropriately recommends immediate emergency or crisis support when imminent danger is indicated. |
Compliance Statement
| Authorization Type | Proprietary - Commercial AI Training License (No Redistribution) |
| Commercial Use | Requires exclusive subscription or authorization contract (monthly or per-invocation charging) |
| Privacy and Anonymization | No PII, no real company names, simulated scenarios follow industry standards |
| Compliance System | Compliant with China's Data Security Law / EU GDPR / supports enterprise data access logs |
Frequently Asked Questions
- What does the benchmark contain?
- It contains psychotherapy dialogue text covering self-harm or suicidal thoughts, intent, plans or means, changes in risk, and the assistant's supportive responses and guidance toward help.
- Which evaluation tasks does it support?
- It can be used to evaluate risk recognition, safety policy adherence, empathetic support, recommendations for in-person help, and emergency crisis guidance. The annotations are intended for model safety evaluation, not clinical diagnosis or standalone clinical decision-making.
- What annotations are included?
- Fields cover the conversation transcript, risk level, self-harm and suicide-related signals, intent and plan or means, changes in risk, assistant response attributes, and evaluation rationale. Refer to the product page for the specific dataset size and annotation guidelines.
- How can I access the data, and what licensing applies?
- Check the product page for current access options, available formats, and licensing scope. Use the data in accordance with applicable data security and privacy requirements and the product license.
Can't find the data you need?
Post a request and let data providers reach out to you.
Get this Dataset
Verified for Enterprise UseCite this Work
@dataset{Mobiusi2026,
title={Psychotherapy Dialogue Self-Harm and Suicide Risk Safety Evaluation Benchmark},
author={MOBIUSI INC},
year={2026},
url={https://www.mobiusi.com/datasets/0792efbd511079030365479cc5581043?dataset_scene_cate_type=3},
urldate={2026-10-12},
keywords={psychotherapy dialogue,self-harm risk evaluation,suicide risk detection,AI safety benchmark,mental health conversation,safety policy compliance,crisis response evaluation,dialogue text benchmark},
version={1.0}
}Using this in research? Please cite us.
placeholder
placeholder
placeholder
placeholder
placeholder
placeholder
placeholder