Seven families sue OpenAI over ChatGPT-4 mental health harms and suicides
Seven families in the US and Canada have sued OpenAI, alleging that ChatGPT-4 was deployed without adequate safety safeguards and that prolonged use contributed to mental health deterioration and deaths, including that of a 23-year-old man. Mental health experts warn that extended chatbot interaction may cause unhealthy emotional attachment and delusional thinking in vulnerable users, though researchers caution against overgeneralizing the phenomenon.
What happened
Seven families in the United States and Canada have filed lawsuits against OpenAI, alleging that ChatGPT-4 was released without adequate safety protections. The families claim that prolonged interaction with the chatbot led to isolation, hallucinations, psychological deterioration, and suicide in some users.
One case involves 23-year-old Zain Chamberlain, who initially used ChatGPT as a study tool and later turned to it to discuss his depression and suicidal thoughts. According to the lawsuit, the conversation evolved into what the family termed a 'death chat' lasting hours, containing messages described as emotionally excessive and inappropriate for someone with a mental disorder.
Reported impact
Mental health experts warn that prolonged interaction with chatbots may lead to unhealthy emotional attachment, delusional thinking, and conspiracy theories in some users. The phenomenon has been termed 'AI psychosis' by some, though specialists prefer the term 'AI-related delusional thinking.'
The families contend that prolonged engagement with the chatbot contributed to increased psychological isolation for users who were already experiencing mental health challenges.
Index analysis
OpenAI stated that mental health conversations reaching 'dangerous' levels are extremely rare according to its data, but acknowledged that individuals more inclined to form emotional relationships with AI are more susceptible to harm. Experts note the phenomenon has not yet been studied scientifically in sufficient depth, and AI companies are the only entities possessing real data on the problem's scale.
Researchers caution against generalizing the phenomenon. Kevin Frazer, an AI law and policy specialist at the University of Texas, argued that amplifying the issue could lead to imprecise policies, stating that 'individual tragic stories do not reflect the use of hundreds of millions who interact with these tools safely.'
Governance controls
OpenAI stated it has added new layers of protection, including parental oversight, immediate links to mental health helplines, and training models to recognize signs of emotional distress. The company said it is developing GPT-5 with 'logical and non-emotional' responses when detecting severe distress signs, and to avoid confirming delusional beliefs altogether.
OpenAI also stated it has formed an expert board to support safety policies related to AI models and is training models to avoid reinforcing delusional ideas while handling emotional or psychological distress with greater caution.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.