John Torous

Overview

John Torous is the psychiatrist interviewed by Marketplace Tech about emotional support, mental-health information, long-term attachment, and clinical evidence around general-purpose AI chatbots.

Current Profile

In the bounded source, Torous’s main contribution is a two-level safety model. Some failures are immediately visible in a dangerous answer, while others emerge through a long relationship in which memory, agreeableness, emotional intensity, and perceived authority accumulate. He therefore argues for Longitudinal Chatbot Safety Evaluation grounded in real user experience and for more precise clinical reporting than the umbrella label AI Psychosis.

His position is neither blanket endorsement nor blanket rejection. General-purpose chatbots are not presented as clinical tools, but basic psychoeducation, wellness information, or emotional support may have value for some users. Safety depends on recognizing warning signs, preserving human relationships, disclosing AI use to treatment providers, and distinguishing conversational reinforcement from the causes and diagnoses of mental illness.

Key Characteristics

• Distinguishes immediately unsafe output from cumulative relationship-level harm.
• Prioritizes user-created, longitudinal tests over model-only evaluation pipelines.
• Treats agreeableness as a safety tradeoff when correction or escalation is needed.
• Calls for tool identity, logs, psychiatric history, and physical-state evidence in reported crises.
• Preserves a limited wellness role for AI while rejecting general-purpose chatbots as clinical authorities.

Evidence

• Harm taxonomy - The promise and risks of AI for mental health support attributes to Torous the distinction between obvious dangerous answers and harm emerging over thousands of interactions.
• Evaluation practice - The promise and risks of AI for mental health support links him to MindBench.AI and its use of real AI users to construct test cases.
• Clinical precision - The promise and risks of AI for mental health support records his argument that AI-themed delusions, pre-existing illness, sleep deprivation, delirium, and chatbot reinforcement require case-specific evidence.
• Practical boundary - The promise and risks of AI for mental health support presents his warning signs and recommendation to keep trusted people and clinicians informed.

Qualifications

This profile rests on one short interview summarized in a supplied episode note, not Torous’s full publication record, clinical guidance, or a verbatim transcript. The source does not provide study methods for its prevalence estimate, comparative model results, or outcome data. Its distinctions are useful evaluation and reporting frames, not individualized diagnosis or treatment advice.

What Changed

• Created a profile for the episode’s central psychiatric expert.
• Captured his distinction between output-level and relationship-level chatbot harm.
• Preserved his cautious account of both limited benefit and unresolved clinical risk.

Relationships

• Marketplace Tech - interview venue for the bounded account.
• MindBench.AI - user-centered evaluation project associated with his discussion.
• Longitudinal Chatbot Safety Evaluation - evaluation approach implied by his harm taxonomy.
• AI Psychosis - broad public label he argues should be clinically disaggregated.
• Human Judgment Under AI - human-check-in and clinical-disclosure boundary reinforced by his guidance.