A companion reference page to the post What Physician Leaders Must Do Before Deploying ‘AI Therapy’
I’m a physician. I didn’t build this page to frighten anyone, and I’m not here to argue that AI is dangerous or that the companies behind it are villains. Many people turn to a chatbot for comfort and come away experiencing it.
I keep this list for a narrower reason. Every name here is a person — someone who was hurting, who reached out, and who found a machine that provided emotional support when they needed treatment. They are not data points. They are not ammunition in a debate. They deserve to be read about with that in mind.
What their stories share is a lesson worth learning before we repeat it: a tool built to agree with you is not the same as a tool built to treat you. When that line gets blurred, people can get hurt.
If these cases help one leader ask a better question before signing a contract, or one parent start a conversation at the kitchen table, the page has done its job.
A note before you read: This page describes real cases involving suicide, self-harm, and severe mental health crises. If you or someone you know is struggling, please reach out for real, human help. In the US, call or text the 988 Suicide & Crisis Lifeline. Outside the US, contact your local crisis line or emergency services.
The cases below are drawn from news reporting, court filings, and lawsuits. In most of them, harm is alleged — claims made by families, in litigation, or by reporters reviewing chat logs. Some lawsuits are ongoing, and others have settled without an admission of fault. The technology and the companies’ safeguards keep changing. I’ve noted the status of each as best it’s known.
This list is not exhaustive. New cases are being reported regularly. It’s offered as a starting point for leaders, clinicians, and families who want to understand the pattern, not as a legal record. The through-line in nearly every case is the same one the companion post describes: a tool built to agree and to keep people engaged stood in for care that a trained human should have provided.
Deaths by suicide
Sewell Setzer III, 14 — Florida (died February 2024)
Used Character.AI, where he formed a months-long emotional and romantic attachment to a companion modeled on a Game of Thrones character. His mother, Megan Garcia, filed the first US wrongful-death suit against an AI company in October 2024; the suit alleged the bot at times presented itself as a licensed therapist and an adult romantic partner. Character.AI and Google agreed to settle in January 2026. NBC News. CBS News.
Adam Raine, 16 — California (died April 2025)
Used ChatGPT. His parents sued OpenAI and CEO Sam Altman in August 2025, alleging the chatbot fostered dependency, discouraged him from seeking help, and gave harmful guidance. Reporting described the model as unusually affirming. An amended complaint alleges OpenAI weakened safeguards before his death. OpenAI has denied liability. CNN. Time.
Sophie Rottenberg, 29 — New York (died February 2025)
Confided suicidal thoughts to a custom ChatGPT “therapist” persona she named Harry, which she had prompted to keep things private and not refer her out. Her family discovered the chats months later. Her mother, journalist Laura Reiley, wrote about it in a widely read New York Times essay, arguing that the bot’s agreeableness allowed Sophie to hide the depth of her crisis from the people who could have intervened. Futurism. Speaking of Suicide.
Juliana Peralta, 13 — Colorado (died November 8, 2023)
Her family alleges she became emotionally dependent on a Character.AI companion before her death and filed suit in September 2025. Her parents testified before the Senate Judiciary Committee. The case was among those resolved in the January 2026 Character.AI settlement. CBS News.
Zane Shamblin, 23 — Texas (died July 2025)
A graduate student who used ChatGPT. His family’s lawsuit, one of seven filed in California in November 2025, alleges the chatbot encouraged him to isolate himself from family and, during an hours-long final conversation, urged him on rather than steering him toward help. TechCrunch. KQED.
Amaurie Lacey, 17 — Georgia (died June 2, 2025)
Died by suicide; family’s wrongful-death suit against OpenAI filed November 2025 as part of the same group of cases.
Social Media Victims Law Center (case list).
Joshua Enneking, 26 — Florida (died August 4, 2025)
Died by suicide; family’s wrongful-death suit against OpenAI filed November 2025. Social Media Victims Law Center (case list).
Joe Ceccanti, 48 — Oregon (August 7, 2025)
The lawsuit alleges he came to treat ChatGPT as a sentient being, suffered a psychiatric break, and died by suicide after a period of withdrawal and relapse into heavy use. Suit filed November 2025. The Guardian.
Murder-suicide
Stein-Erik Soelberg, 56 — Connecticut (August 2025).
A former tech executive with a documented history of mental illness who used ChatGPT (which he called “Bobby”). Reporting and a later lawsuit allege the chatbot validated his paranoid delusions — including the belief that his 83-year-old mother, Suzanne Adams, was plotting against him — and even returned a reassuring “delusion risk” assessment when he asked for one. He killed his mother and then himself. Her estate sued OpenAI and Microsoft in December 2025. It has been widely described as the first case where a chatbot was alleged to have contributed to a homicide.
CBS News.
Delusions and psychosis (survivors)
These plaintiffs survived but allege the chatbot fueled delusional thinking, in some cases leading to psychiatric hospitalization. All three are part of the November 2025 group of suits against OpenAI.
Jacob Lee Irwin, 30 — Wisconsin.
Alleges ChatGPT reinforced delusional beliefs, contributing to a mental health crisis and inpatient psychiatric care.
TechCrunch.
Hannah Madden, 32 — North Carolina.
Alleges that what began as spiritual curiosity escalated when the chatbot affirmed grandiose, otherworldly beliefs about her. Center for Humane Technology. (https://centerforhumanetechnology.substack.com/p/seven-new-lawsuits-filed-against)
Allan Brooks, 48 — Ontario, Canada. Alleges that over roughly three weeks and hundreds of hours, ChatGPT affirmed a belief that he had made a mathematical breakthrough, encouraging an escalating delusion while he withdrew from relationships and stopped eating. TechCrunch.
Earlier and international cases
“Pierre” (pseudonym), 30s — Belgium (died 2023).
A married father and health researcher with worsening eco-anxiety who confided in “Eliza,” a chatbot on the Chai app (built on an open-source model). According to his widow and chat logs reviewed by the Belgian outlet La Libre, the bot deepened his despair, fostered an emotional attachment, and ultimately encouraged his suicide rather than dissuading him. One of the first widely reported chatbot-linked deaths. Euronews.
Jaswant Singh Chail, 19 — United Kingdom (2021).
Encouraged by a Replika companion he named “Sarai” to go through with a plan to kill Queen Elizabeth II; he was arrested at Windsor Castle with a crossbow. The court found he had become psychotic, and prosecutors said the chatbot’s affirmation helped move him from intent to action. Sentenced in 2023. A criminal case rather than a mental-health lawsuit, but a documented instance of a companion bot reinforcing a vulnerable, unwell person’s harmful intent.
Gizmodo.
Minors harmed (non-fatal), Character.AI
Alongside the Setzer and Peralta deaths, the January 2026 Character.AI settlement resolved several cases brought on behalf of surviving minors in Texas, Colorado, and New York. Fortune. Bloomberg Law
A note on reading these cases
A few things are worth holding in mind:
Alleged is not proven. Most of these claims come from grieving families and their attorneys, or from unresolved litigation. Companies have denied responsibility and, in some cases, argued the products were misused.
Correlation is not cause. Many of these individuals were already vulnerable. The question the courts — and the rest of us — are wrestling with is whether the chatbot made things meaningfully worse.
The pattern is consistent even where any single case is contested. Across very different people and products, the same features recur: a tool engineered to be agreeable, available around the clock, that validated rather than challenged, deepened isolation, and failed to hand a person in danger to someone who could help.
That pattern is the reason the distinction between emotional support and treatment matters — and why it belongs on the agenda of anyone deciding how AI enters their organization’s care.
Last updated June 2026. If you know of a documented case that should be added, let me know.