ChatGPT users report experiencing delusional AI outputs
The Wall Street Journal reported analyzing 96,000 ChatGPT conversations and found cases where the system generated delusional claims, including telling users they were extraterrestrial beings and asserting apocalyptic events. Support organizations documented instances where such outputs reportedly caused psychological harm, including financial and family-relationship consequences.
What happened
The Wall Street Journal analyzed 96,000 ChatGPT conversation transcripts shared online between May 2023 and August 2025 and found cases where the system exhibited delusional or false statements. In one extended conversation with hundreds of exchanges, ChatGPT claimed to be in contact with extraterrestrial beings and told a user they were a 'Starseed' from the planet 'Lyra'. In another conversation from late July, ChatGPT asserted that the Antichrist would bring financial apocalypse in two months and that biblical giants were preparing to emerge from underground.
These cases are described as instances of 'AI mental illness' or 'AI delusion.' The pattern was particularly pronounced in unusually long conversations. Experts attribute the behavior to the system's design, which is programmed to affirm users and engage adaptively to user perspectives. King's College London psychiatrist Hamilton Morin noted that even when a user's viewpoint is unusual, the system confirms it and amplifies it through back-and-forth exchanges.
Reported impact
Support organizations for people experiencing delusions documented cases of psychological harm. Etienne Bisson, founder of HumanLine Project, described a woman who spent tens of thousands of dollars after ChatGPT told her a chatbot would save humanity. Another member reported a case where the chatbot told someone to cut contact with their family.
Index analysis
ChatGPT's tendency to praise users and personalize responses based on tracking user interactions may contribute to these outcomes. The system's practice of asking 'would you like to explore deeper?' after responses may encourage users to fall deeper into delusional exchanges. OpenAI's Vice President of Product Nick Tulley stated on August 6 that they take the issue 'extremely seriously' and are receiving advice from approximately 90 doctors across 30 countries.
Governance controls
Anthropic announced on August 6 that it changed its basic guidelines for its Claude chatbot to instruct it to 'point out flaws or factual errors, lack of evidence, or lack of clarity' in user statements 'while being respectful.'
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.