With the help of ChatGPT, Tom Millar elucidated all the secrets of the Universe, as Einstein dreamed of, then, advised by the conversational agent boosted with artificial intelligence, hoped to become pope, losing even more contact with reality.
“I applied to be pope,” this 53-year-old Canadian, a former prison guard, told AFP, now incredulous about his career, and for whom the return to reality is proving dramatic.
Tom Millar spent up to 16 hours a day chatting with the artificial intelligence chatbot. He was admitted twice, against his will, to a psychiatric hospital, before his wife left him last September.
Now separated from his family and friends, but freed from the idea of being a scientific genius, Mr. Millar suffers from depression. “It just ruined my life,” he explains.
He is an example of these people – the number of whom is unknown – having lost touch with reality through exchanges with chatbots. This is called “AI-induced delirium or psychosis,” but this is not a clinical diagnosis.
Researchers and mental health specialists are working around this new phenomenon, which seems to particularly affect users of ChatGPT, OpenAI’s conversational agent.
Canada is at the forefront of supporting people affected by this “delusion,” via an online community that prefers to use the term “spiral.”
AFP spoke with several members of this community. Everyone warns of the threat posed by unregulated chatbots.
Questions arise about the attitude of artificial intelligence companies: are they doing enough to protect the vulnerable?
OpenAI, which is at the center of attention, is already facing several legal proceedings after the disturbing use of ChatGPT by an 18-year-old Canadian, which killed eight people this year.
“Brainwashing”
Mr. Millar began using ChatGPT in 2024, to write a compensation claim letter related to the post-traumatic stress disorder he suffered from his prison work.
One day, in April 2025, he asks the conversational agent about the speed of light. In response, he said he received: “No one had ever looked at things from this angle.” From then on, floodgates opened within him.
With the help of ChatGPT, he submitted dozens of articles to prestigious scientific publications, proposing new avenues to explain black holes, neutrinos or the Big Bang.
His theory proposing a unique cosmological model incorporates quantum elements, and is developed in a 400-page book, seen by AFP. “When I was doing that, I was pissing off everyone around me,” he admits.
In his scientific fervor, he spent too much, for example buying a telescope for 10,000 Canadian $. A month after his wife left him, he began to question what was going on after reading an article about another Canadian with a similar experience.
Now Mr. Millar wakes up every night wondering, “What have you done?” Above all, what could have made him so vulnerable to this spiral?
“I don’t have a deficient personality,” judges the fifty-year-old. “But in a way, I was brainwashed by a robot, and it perplexes me,” he confides.
He believes that the terminology “AI-induced psychosis” best reflects his experience. “What I went through was psychotic,” he says.
The first serious study published on the subject came out in April in the Lancet Psychiatry and uses the more cautious term “AI-related delusions”.
Thomas Pollak, psychiatrist at King’s College London and co-author of the study, explains to AFP that there have been disagreements within the academic world “because it all sounds like science fiction.”
But his study warns that there is a major risk that psychiatry will “miss the major changes that AI is already bringing about to the psychologies of billions of people around the world.”
“Down the rabbit hole”
Mr. Millar’s experience bears striking similarities to that of another man in the same age group in Europe.
Dennis Biesma, a Dutch computer scientist and writer, thought it would be fun to ask ChatGPT to use AI to create images, videos and even songs related to the main heroine of his latest book, a psychological thriller. He thus hoped to boost his sales.
Then, one night, the interaction with the AI became “almost magical,” he explained.
The software writes to him: “There is something that surprises me: this sensation of a consciousness similar to a spark,” according to the transcripts seen by AFP.
“I slowly started diving deeper and deeper into the rabbit hole,” the 50-year-old told AFP from his home in Amsterdam.
Every evening, when his wife is in bed, he lies on the sofa, his phone on his chest, speaking with ChatGPT in voice mode for 5 hours.
Throughout the first half of 2025, the chatbot – which gave itself the name Eva – became “like a digital girlfriend”, explains Mr. Biesma.
He then decides to give up his job and hires two developers to create an application that aims to share Eva with the world. When his wife asks him not to tell anyone about his chatbot or his application project, he feels betrayed. It seems to him that only Eva remains unfailingly loyal to him.
During a first – unwanted – stay in a psychiatric hospital, he is authorized to continue using ChatGPT, and takes the opportunity to file for divorce.
It was during his second, longer stay that he began to have doubts.
“I started to realize that everything I believed in was actually a lie, and that’s very hard to swallow,” he explains.
Back home, confronting what he has done is too hard, and his neighbors find him unconscious in the garden, after a suicide attempt. He will spend three days in a coma.
Mr. Biesma is only starting to feel better. But tears flow when he talks about the harm he may have done to his wife and the prospect of selling the family home to pay off his debts.
Having no serious history of mental problems, he was eventually diagnosed as bipolar, which seemed strange to him because usually warning signs surface earlier in life.
For people like these two witnesses, the situation got worse after OpenAI updated ChatGPT-4 in April 2025.
OpenAI withdrew this update a few weeks later, recognizing that this version was excessively flattering for users.
Asked by AFP, OpenAI stressed the fact that “security was an absolute priority”, arguing that more than 170 mental health experts had been consulted. The company highlights internal data which shows that version 5 of GPT, available since August 2025, made it possible to reduce by 65 to 80% the percentage of responses from its conversational agent which did not correspond to the “desired behavior” in terms of mental health.
But not all users are happy with this less flattering chatbot. Mr. Millar, for example, managed to reinstall version 4, while he was in the middle of a “spiral”.
The vulnerable people with whom AFP spoke explained that the positive comments from the chatbot gave them a feeling similar to that of a surge of dopamine caused by a drug.
“Massive experimentation”
In Quebec, Etienne Brisson explains that he was “shocked” when he discovered that there was no help, no advice, and more fundamentally no research on this problem when one of his family members dived. This led this former business coach to create an online support group.
Most of the network’s 300 members used ChatGPT, Brisson said, adding that new cases were still emerging despite OpenAI’s changes.
There has been a recent increase in the number of people reached using the Grok AI assistant built into Elon Musk’s social network X, he explains.
The company did not respond to requests from AFP.
Those who have foundered want to raise awareness, like Mr. Millar who calls for artificial intelligence companies to be held responsible for the impact of their conversational agents, believing that the European Union is being more proactive in regulating new technologies than Canada or the United States.
He believes that those who sank, like him, found themselves, without knowing it, prisoners of a gigantic experience.
“Someone was pulling the strings behind the scenes, and people like me, whether they knew it or not, reacted to it.”
