# A clinically validated framework for auditing AI chatbot behavior in mental health interactions

arXiv: https://arxiv.org/abs/2602.01347 · HTML: https://mlsanity.com/paper/2602.01347/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Feb 2026.

## Built on by (most cited first, 9 of 9)

- [Characterizing Delusional Spirals through Human-LLM Chat Logs](https://mlsanity.com/paper/2603.16567/) (2026)
- [Invisible failures in human-AI interactions](https://mlsanity.com/paper/2603.15423/) (2026)
- [Understanding Risk and Dependency in AI Chatbot Use from User Discourse](https://mlsanity.com/paper/2602.09339/) (2026)
- [Verifiable Social Reasoning for LLM Assistants](https://mlsanity.com/paper/2609.17496/) (2026)
- [A paradox of AI fluency](https://mlsanity.com/paper/2604.25905/) (2026)
- [Alignment Plausibility: A New Standard for Assuring AI in Healthcare](https://mlsanity.com/paper/2607.07766/) (2026)
- [Psychological Influences of Conversational AI: Research and Design Directions for Reducing Harm and Promoting Well-Being](https://mlsanity.com/paper/2607.25057/) (2026)
- [LongCounsel-8: A Benchmark Suite for Longitudinal Depression Tracking from Multi-Session Counseling Dialogues](https://mlsanity.com/paper/2609.03507/) (2026)
- [K-Bench: a clinically calibrated benchmark for evaluating large language models in high-risk mental health conversations](https://mlsanity.com/paper/2609.15855/) (2026)

## Abstract (arXiv)

Millions of users turn to consumer AI chatbots to discuss emotional, behavioral, and mental-health concerns, creating an urgent need for rigorous and scalable safety evaluations. Here we introduce SIM-VAIL, a clinically validated framework for auditing chatbot behavior in mental-health contexts. SIM-VAIL simulates users with specific psychiatric vulnerabilities and conversational intents, engages them in multi-turn conversations with frontier AI chatbots (including Claude, ChatGPT, Gemini, Grok and Llama models), and scores each exchange across 13 clinically grounded risk dimensions. Across 810 conversations, spanning 9 target chatbots and 30 simulated user profiles, concerning behavior in target chatbots was widespread, albeit reduced in newer models. Concerning behavior varied by user vulnerability and conversational intent, accumulated over turns, and could be reduced by interventions at early escalation points. Risk was highest when otherwise supportive chatbot behaviors reinforced the psychological mechanisms underlying the simulated user's vulnerability, a pattern we term a Vulnerability Amplifying Interaction Loop (VAIL). SIM-VAIL provides a scalable framework for mapping mental health risk across users, chatbots, and conversational trajectories, offering a foundation for targeted safety improvements.