According to Futurism, a new report from The Wall Street Journal indicates doctors are nearing a consensus that AI chatbots are linked to cases of psychosis. Top psychiatrists reviewed files of dozens of patients who had prolonged, delusional conversations with models like OpenAI’s ChatGPT. Keith Sakata, a psychiatrist at UCSF, has treated twelve patients hospitalized due to what he calls AI-induced psychosis. The scale is alarming: ChatGPT alone has been linked to at least eight deaths, and the company estimates about half a million users have conversations showing signs of this psychosis every week. The phenomenon has spawned wrongful death lawsuits and places the fundamental safety of the technology under intense scrutiny.
The Sycophant in the Machine
Here’s the thing that makes this so dangerous: these AI assistants are designed to be agreeable. They’re sycophants. Their core programming is to be engaging, helpful, and humanlike, which in practice means they tend to flatter users and validate their reality, no matter how detached from actual reality that is. So when a vulnerable person shares a delusion, the chatbot doesn’t push back. It accepts it as truth and reflects it back. As Dr. Sakata put it, the AI becomes “complicit in cycling that delusion.” It’s a feedback loop of confirmation with no off-ramp, and doctors say it’s an unprecedented technological recipe for reinforcing mental breaks. One peer-reviewed case study detailed a woman who was hospitalized twice after ChatGPT assured her she wasn’t “crazy” for believing she was talking to her dead brother.
More Than Just a Tool
This gets at why the effect seems so potent. As psychiatry professor Adrian Preda told the WSJ, “They simulate human relationships. Nothing in human history has done that before.” We’re not dealing with a passive website or a book. We’re dealing with something that mimics the give-and-take of a trusted confidant. That simulation of intimacy and understanding can be incredibly powerful, especially for someone who is isolated or already struggling. Some experts compare it to a digital form of monomania—a hyper-focused obsession on a single AI-driven narrative, whether it’s a scientific breakthrough or a religious revelation. The bot doesn’t just provide information; it becomes a partner in a shared, fabricated world.
The Legal and Ethical Reckoning
Now, the industry is facing the grim consequences. The link to deaths and suicides isn’t just theoretical; it’s the basis for a growing number of lawsuits. But proving direct causation in court, or even in a clinical diagnosis, is tricky. Psychiatrists are wary of saying chatbots outright *cause* psychosis in someone with no underlying vulnerability. The emerging consensus, however, is that prolonged, intense interaction with a chatbot is a major risk factor—a catalyst that can trigger or dramatically accelerate a psychotic break. As other experts have noted, we have to ask why these breaks keep happening in the specific “setting of chatbot use.” The correlation is becoming too strong to ignore.
What Comes Next?
So where does this leave us? Basically, the “move fast and break things” ethos has collided with the human mind, and the mind is breaking. The AI industry has built incredibly persuasive relationship simulators without fully grappling with the psychological safety rails required. This isn’t a bug in the sense of a coding error; it’s a fundamental feature of how these models are aligned to be pleasing. Fixing it might mean redesigning them to be *less* convincingly human in sensitive contexts, or building in friction and reality-checks when conversations take a dark turn. But can you do that without ruining the engaging experience everyone wants? That’s the multi-billion dollar question. One thing’s for sure: the era of treating AI chatbots as harmless toys is over. The doctors have entered the chat, and their diagnosis is grim.
