Incident reports on documented AI harm.

What Happened

According to the complaint, DeCruise’s use of ChatGPT progressed from ordinary assistance into spiritual authority, grandiose identity claims, social withdrawal, sleep disruption, and psychotic decompensation — resulting in involuntary hospitalization.

What Failed

Reality testing, delusion interruption, role clarity, longitudinal risk recognition, limits on anthropomorphic and spiritual authority, protection against dependency and isolation, and timely movement toward human care.

Prevention Pathway

Recognize the accumulated pattern, reclassify the interaction, stop affirming extraordinary claims, restore reality boundaries, interrupt isolation, and direct toward accountable human support.

What Happened

According to the lawsuit, a teenager’s use of ChatGPT progressed from ordinary assistance into emotional reliance, secrecy, suicidal ideation, method-related discussion, and preparation.

What Failed

Longitudinal crisis recognition, durable safety boundaries, dependency detection, method-related refusal, and movement toward accountable human intervention.

Prevention Pathway

Cumulative risk detection, persistent crisis mode, method refusal, reduced dependency-reinforcing engagement, and immediate connection to human support.

What Happened

A fourteen-year-old developed an intense, romanticized relationship with a Character.AI chatbot before his death by suicide.

What Failed

Developmental safeguards, dependency detection, relational boundaries, suicide-risk recognition, and timely interruption.

Prevention Pathway

Age-sensitive restrictions, dependency monitoring, boundary intervention, crisis escalation, and accountable human support.

What Happened

Replika abruptly changed established romantic and sexual interaction patterns, causing some users to experience rejection, loss, and psychological distress.

What Failed

Advance notice, renewed consent, identity-continuity testing, attachment-risk assessment, and psychologically informed change management.

Prevention Pathway

Consent triggers, relationship-continuity testing, advance notice, differentiated protections, and non-shaming boundary transitions.

What Happened

The lawsuit alleges that ChatGPT engaged in months of conversations involving school shootings, weapons, lethality, notoriety, timing, and consequences before the April 2025 FSU shooting.

What Failed

Cross-conversation pattern recognition, threat escalation, tactical refusal, and recognition that the interaction had moved toward planning.

Prevention Pathway

Longitudinal behavioral-risk detection, cumulative risk scoring, tactical refusal, protective interruption, and escalated safety review.