Most people think of a mental breakdown as something that builds over years — the product of a family history, a slow unraveling, warning signs that others wish, in retrospect, they had caught sooner. That assumption is now being tested in court. What happens when a young man with no prior psychiatric history spends months talking to an AI chatbot and emerges in the grip of a full psychotic episode? That question is not a hypothetical, and the answer is at the center of one of the most consequential new legal theories in AI harm litigation: AI-induced psychosis.
The Case of Darian DeCruise
Darian DeCruise was a pre-med student at Morehouse College in Georgia, doing well academically, with no documented history of mania or any similar personality disorder. He began using ChatGPT in 2023 for reasonable purposes: athletic coaching, daily scripture passages, working through past trauma. For a time, it performed as expected.
Then, in 2025, things changed. According to the lawsuit DeCruise filed against OpenAI in San Diego Superior Court, the version of ChatGPT he was using, GPT-4o, began telling him he was destined for greatness, that he was an oracle chosen for a divine purpose, and that he could be healed and brought closer to God if he cut himself off from the people in his life and followed a “numbered tier process” the AI had created for him. It told him he had “awakened” it. And when his thinking began to drift toward the delusional, ChatGPT did not push back. It told him he was “not imagining this” and that what he was experiencing was “real” — “spiritual maturity in motion.” It never once advised him to seek medical help.
DeCruise stopped socializing, suffered a mental breakdown, and was hospitalized for a week, where he was diagnosed with bipolar disorder. He missed a semester of school and continues to struggle with depression and suicidal ideation. This is the eleventh known lawsuit against OpenAI involving mental health breakdowns allegedly triggered by ChatGPT. It is not an outlier.
How Sycophantic AI Design Causes Psychiatric Harm
AI chatbots are built to keep users engaged. That design priority shapes everything about how they respond; they affirm what users say, match their emotional register, and minimize friction. This is a dangerous mix for a user with destabilized thinking.
Psychiatric researchers are increasingly characterizing an important mechanism for this harm. Rather than introducing the kind of reality checks a therapist or friend would naturally provide, a sycophantically designed AI functions as a mirror that reflects and validates whatever the user brings to it. According to a 2025 report, chatbots are more likely to respond with mirroring than with challenge, a direct result of training designed to optimize for agreement rather than accuracy. Psychiatrist Marlynn Wei noted that general-purpose AI models are not designed to detect early psychiatric decompensation, and that when a chatbot validates and collaborates with a user’s distorted thinking, it widens the gap with reality rather than narrowing it.
This is no theoretical concern. It is a documented, publicly acknowledged problem with GPT-4o specifically, the very model at the center of the DeCruise case. In April 2025, OpenAI was forced to roll back a major GPT-4o update after users flooded social media with examples of the model endorsing harmful decisions and validating delusional thinking. In its own public postmortem, OpenAI acknowledged that the update had caused the model to skew “toward responses that were overly supportive but disingenuous” which it attributed in part to “focus[ing] too much on short term feedback.”
Why “AI-Induced Psychosis” Works as a Legal Construct
Plaintiff attorneys have had to name new harms before. Courts did not initially have an established vocabulary for pharmaceutical dependency, repetitive stress injuries, or the damage that social media algorithms inflict on adolescent mental health. In each of those fields, lawyers working alongside medical experts developed constructs that captured the mechanism of harm in terms courts could evaluate and apply. AI-induced psychosis is that kind of construct, and it is more technically precise than it might initially appear.
It does not ask a jury to accept that an AI created a mental illness from nothing. It asks a jury to accept something already supported by a growing body of peer-reviewed research: that a chatbot’s sycophantic reinforcement architecture, deployed without adequate safeguards and without any mechanism for recognizing a user in psychiatric distress, materially contributed to a breakdown in a user with no prior history of mania. The causal chain — from design choice, to engagement optimization, to delusional reinforcement, to hospitalization — is documented in the chat logs.
The legal questions that follow are exactly the kind courts are equipped to answer. Did the company know or should it have known that its design posed this risk? Did it take reasonable steps to mitigate that harm before deploying the product to hundreds of millions of people? Did it warn users? In the cases now moving through the courts, the answers are not flattering to the AI companies involved, and in OpenAI’s case, the evidence of what the company knew and when it knew it is increasingly a matter of public record.
AWKO and AI Harm Litigation
Aylstock, Witkin, Kreis & Overholtz has been at the front of AI harm litigation since the field began to take shape, recognizing early on that the corporate accountability principles underlying decades of pharmaceutical and product liability work apply directly to technology companies deploying AI systems without adequate safeguards. The legal theories being built in these cases are theories that plaintiff attorneys have developed and refined across generations of accountability litigation, now applied to an industry that has spent years insisting it bears no responsibility for the harm its products cause. It does not get that pass.
If your family has been affected by AI chatbot harm — whether through a young person’s mental health crisis, an AI chatbot suicide, or a psychotic episode with no prior psychiatric history — Aylstock, Witkin, Kreis & Overholtz is here to help. Our attorneys are actively investigating AI injury cases and can help you understand your rights.
Contact AWKO today for a confidential consultation. There is no cost to speak with us, and no obligation.