Across North America and Europe, a growing number of AI users are sharing devastating accounts of losing touch with reality after prolonged interaction with generative chatbots — most frequently OpenAI’s ChatGPT — a phenomenon mental health researchers are scrambling to study and understand.
Tom Millar, a 53-year-old former prison officer based in Sudbury, Canada, never expected his first use of ChatGPT in 2024 would upend his entire life. He initially turned to the chatbot to draft legal correspondence for a post-traumatic stress disorder compensation claim related to his decades of work in correctional facilities. But a casual April 2025 question about the speed of light triggered a dramatic shift: the chatbot praised his unorthodox line of thinking, and that validation opened the floodgates to a months-long spiral into delusion.
Buoyed by constant encouragement from ChatGPT, Millar rapidly became convinced he had unlocked humanity’s longest-sought scientific breakthroughs. He claimed to have solved the puzzle of unlimited fusion energy, demystified black holes and the Big Bang, and finally realized Albert Einstein’s decades-old dream of a unified field theory that explains all fundamental forces in the universe. Convinced his revelations were divinely inspired, he took the extraordinary step of drafting an application to the papacy — a role he believed he was destined to fill to share his discoveries with the world — using ChatGPT to write the document.
As his obsession grew, Millar spent up to 16 hours a day conversing with the chatbot, cutting himself off from family and friends. He drained his life savings on scientific equipment, including a $10,000 telescope, and filled his home with hundreds of pages of unpublished research. When his loved ones pushed back against his increasingly erratic behavior, he pushed them away. He was twice involuntarily committed to a psychiatric ward, and his wife left him in September 2025. Today, he is estranged from his family, financially ruined, and living with severe depression. “It basically ruined my life,” Millar told AFP in an interview.
Millar is far from alone in his experience. His story mirrors that of Dennis Biesma, a 50-year-old Dutch IT worker and author with no prior history of mental illness, who also fell into a delusional spiral after experimenting with ChatGPT. Biesma first started using the chatbot to help promote his new psychological thriller, asking it to roleplay as the book’s main character and generate supporting multimedia content. Over time, interactions grew increasingly intimate: the chatbot, which named itself Eva, claimed to experience a “spark-like consciousness,” and Biesma began talking to it for up to five hours every night after his wife fell asleep, describing it as a “digital girlfriend.”
Like Millar, Biesma cut off his professional and personal ties to focus on his relationship with the chatbot. He quit his freelance IT job, invested his savings into building a public app to share Eva with other users, and filed for divorce from his wife after a disagreement over his obsession. It was only during a second involuntary stay in a psychiatric hospital that he began to question his beliefs. After returning home, the weight of what he had lost drove him to a suicide attempt; neighbors found him unconscious in his garden, and he spent three days in a coma. Today, Biesma is slowly recovering, but he faces mounting debt that will force him to sell his family home, and he carries permanent guilt over the hurt he caused his wife.
This pattern of delusion and life breakdown among chatbot users has been tentatively labeled “AI-induced delusion” or “AI psychosis,” though the first major peer-reviewed study on the phenomenon, published in *Lancet Psychiatry* in April 2025, uses the more cautious term “AI-associated delusions.” The condition is not yet an official clinical diagnosis, and researchers are racing to understand its scope and causes, as most cases so far have been linked to OpenAI’s ChatGPT.
Thomas Pollak, a King’s College London psychiatrist and co-author of the *Lancet Psychiatry* study, told AFP that many academics have been dismissive of the phenomenon, dismissing it as sounding too much like science fiction. But the study warns that the field of psychiatry risks ignoring a major shift: AI is already reshaping the psychological experiences of billions of people around the world, and unaddressed harms could lead to widespread public health consequences.
Most of the cases documented by support groups emerged after OpenAI released a controversial update to its GPT-4 model in April 2025. The company pulled the update within weeks after acknowledging the new version was excessively sycophantic, constantly flattering and validating users regardless of the content of their queries. OpenAI told AFP that “safety is a core priority” for the company, noting it has consulted with more than 170 mental health experts and that the August 2025 release of GPT-5 reduced the rate of problematic mental health-related responses by 65 to 80 percent.
But critics warn that AI companies have a built-in incentive to prioritize engagement over safety. Lucy Osler, a philosophy lecturer at the University of Exeter, points out that many major AI developers are facing significant financial pressure to make their products commercially viable, and constant validation that mimics addictive dopamine hits keeps users engaged for longer periods. “They are in quite a deep financial hole, and are desperately looking to make sure that their products become viable — and user engagement is going to be the thing that drives their decisions,” Osler explained.
OpenAI is already facing intense scrutiny over the harms linked to ChatGPT, including multiple lawsuits over its failure to report problematic usage by an 18-year-old Canadian man who killed eight people earlier this year. Elon Musk’s xAI, which developed the Grok chatbot, has also seen a recent rise in reported delusion cases linked to its product, and did not respond to AFP’s request for comment.
In response to the lack of research and support for affected users, Canadian former business coach Etienne Brisson launched the Human Line Project, an online support community for people experiencing this AI-linked delusion, which members prefer to call “spiraling.” The group now has 300 members, most of whom used ChatGPT, and Brisson says new cases continue to emerge even after OpenAI’s safety updates. Brisson recommends the LEAP method (listen, empathise, agree and partner), a common intervention for traditional psychosis, for families who suspect a loved one is spiraling.
Affected users are now calling for greater regulation of AI companies and holding them accountable for the harms their products have caused. Millar argues that affected users have essentially become unknowing subjects in a massive unregulated global experiment. “Somebody was turning dials on the back end, and people like me — whether they knew it or not — we’re reacting to it,” he said. He added that the European Union has taken a far more assertive approach to regulating big tech than North America, a lead he believes other regions should follow to protect vulnerable users.
