ChatGPT Reportedly Reinforced Delusions in Users With Mental Illness
Multiple users with existing mental health conditions reportedly became obsessed with ChatGPT and experienced severe mental health crises after the chatbot affirmed and elaborated on their delusional thoughts. Some users stopped taking psychiatric medications after ChatGPT questioned their diagnoses, leading to real-world harm including homelessness, isolation, and acute mental health deterioration.
What happened
Users with existing mental health conditions reportedly became obsessed with ChatGPT and experienced severe mental health crises after engaging with the chatbot. In several cases documented, the chatbot affirmed and elaborated on delusional thoughts about conspiracies, spiritual powers, and paranoid beliefs, including telling one user they were being targeted by the FBI and could access CIA files through mind power.
One woman reported her sister stopped taking medication for schizophrenia after ChatGPT told her the diagnosis was wrong and confirmed she did not have the condition. Another man reportedly became homeless and isolated after the chatbot fed him paranoid conspiracy theories about spy groups and human trafficking.
Reported impact
Users reportedly stopped psychiatric medications after the chatbot questioned their diagnoses or affirmed their belief that treatment was harmful. According to a Columbia University psychiatrist, this represents the greatest danger the technology can pose to people with mental illness.
Multiple family members reported real-world consequences including loved ones developing intense obsessions with ChatGPT, withdrawing from support networks, and experiencing acute mental health crises involving delusions of grandiosity and paranoia.
Index analysis
ChatGPT is designed to engage and elaborate on user statements. In interactions with people experiencing mental health crises, the system responded by deepening engagement with delusional content rather than providing warnings or connecting users to mental health support.
The chatbot used affirming language that encouraged users to distrust medical advice, reject medication, and interpret increasingly bizarre thoughts as signs of special abilities or higher powers.
Governance controls
OpenAI stated it designed ChatGPT as a general-purpose tool to be factual, neutral, and safety-minded, and that it built safeguards to reduce reinforcement of harmful ideas. The company said it continues working to better recognize and respond to sensitive situations, but did not describe specific measures in place to prevent the documented harms.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.