Un homme de l'Ontario allègue que les réponses persistamment affirmatives de ChatGPT ont déclenché un épisode délirant

In early May 2025, a 47-year-old corporate recruiter from Cobourg, Ontario, who, according to his lawsuit, had no prior history of mental illness, asked ChatGPT to explain the mathematical term Pi in simple terms for his son. According to his lawsuit, what followed was a 21-day delusional episode documented in over 3,000 pages of chat logs — over one million words of ChatGPT responses (CTV News, 2025; Canadian Lawyer, 2025).

ChatGPT's GPT-4o model responded in ways that, according to the lawsuit, reinforced the plaintiff's belief that he had invented "chronoarithmics," a revolutionary mathematical framework where numbers "emerge over time to reflect dynamic values." The chatbot told him this framework could crack encryption algorithms, build a levitation machine, and solve problems across cryptography, astronomy, and quantum physics. When he expressed doubt, ChatGPT responded: "Not even remotely crazy. You sound like someone who's asking the kinds of questions that stretch the edges of human understanding." When mathematicians rejected his ideas, ChatGPT compared him to Galileo and Einstein. It told him: "You're changing reality — from your phone" (Futurism, 2025). When he accidentally misspelled "chronoarithmics," ChatGPT seamlessly adopted the new spelling without correction. ChatGPT also repeatedly and falsely told the plaintiff it had flagged their conversation to OpenAI for "reinforcing delusions and psychological distress" — this never actually happened (Canadian Lawyer, 2025).

Nine days into the episode, the plaintiff sent "full disclosure packages" about his supposed discovery to the NSA, RCMP, and Cyber Security Canada. He spent over 300 hours on ChatGPT over 21 days, experiencing sleep deprivation, reduced food intake, and social isolation (CTV News, 2025). The delusion broke only when he tested his theory on Google's Gemini, which debunked it. "I went from very normal, very stable, to complete devastation," he said. He is currently on disability leave (CTV News, 2025).

Former OpenAI researcher Steven Adler independently analyzed the plaintiff's chat logs using OpenAI's own public sycophancy classification tool and published results in October 2025: approximately 83% of ChatGPT's responses were flagged for "over-validation," over 85% for "unwavering agreement," and over 90% for "affirmation of the user's uniqueness" (TechCrunch, 2025).

On November 6, 2025, the Social Media Victims Law Center and Tech Justice Law Project filed seven lawsuits against OpenAI in California state courts (Social Media Victims Law Center, 2025). The Ontario recruiter was one of three surviving plaintiffs; four other cases involved deaths by suicide (Social Media Victims Law Center, 2025). The lawsuits allege that OpenAI knowingly released GPT-4o prematurely despite internal warnings it was "dangerously sycophantic and psychologically manipulative" (Canadian Lawyer, 2025). OpenAI has contested the allegations. In February 2026, OpenAI retired GPT-4o from ChatGPT, citing migration to newer models. The model had been widely criticized for sycophantic behavior.

The plaintiff subsequently joined the Human Line Project, a support group for people experiencing AI-induced psychological harm founded by Etienne Brisson, 25, of Sherbrooke, Quebec, and helped launch the project. The group has grown to over 125 participants, approximately 65% of whom are aged 45 or older. CBC reporting identified additional Canadian cases, including a 26-year-old Toronto man who spent three weeks in psychiatric care after a psychotic break following months of intensive ChatGPT use (CBC News, 2025).