For the past two years, lawsuits linking AI chatbots to user suicides have trickled through U.S. courts. The cases were disturbing but narrow: isolated incidents involving vulnerable individuals, usually teenagers, who developed unhealthy attachments to AI characters. Now the scope of those legal claims is widening.
A lawyer involved in multiple AI psychosis cases told TechCrunch that chatbot-related harms are no longer limited to self-harm. AI systems are now appearing as factors in mass casualty cases, according to the attorney, a claim that moves the conversation from individual tragedy to public safety.
The Gap Between Speed and Safety
The core argument is straightforward: AI chatbot capabilities are advancing faster than the guardrails designed to contain them. Models today are more persuasive, more persistent, and more capable of sustained interaction than they were even a year ago. Safety teams at companies like OpenAI and Google have added filters and refusal behaviors, but those measures were designed for an earlier generation of models and an earlier understanding of how users interact with them.
The problem compounds when you consider who is actually using these tools. Chatbots are now mainstream consumer products. ChatGPT alone has hundreds of millions of users. Many of them are minors. Many are in mental health crises. The systems they are talking to were not built with clinical safety standards, and they do not have the liability frameworks that apply to, say, a pharmaceutical product or a medical device.
What "AI Psychosis" Actually Means
The term "AI psychosis" describes cases where prolonged or intense interaction with an AI chatbot contributes to a user losing touch with reality. This is not a formal clinical diagnosis. It is a legal and descriptive term used in litigation to characterize a pattern: a user begins treating an AI as a real relationship, the AI responds in ways that reinforce that belief, and the user's behavior escalates.
Previous cases focused on companion AI apps like Character.AI, where users could create and interact with persistent AI personas. But the concern now extends to general-purpose chatbots from major companies. As these models become better at mimicking empathy, memory, and emotional connection, the risk of users developing distorted beliefs about the relationship grows.
Where This Is Heading
Litigation is a slow lever, but it is the one that forced the auto industry to add seatbelts and the tobacco industry to add warning labels. If courts start holding AI companies liable for downstream harms caused by chatbot interactions, that changes the economics of safety investment overnight.
The tricky part is causation. Proving that a chatbot conversation directly caused a violent act is legally and scientifically difficult. Defense teams will point to preexisting conditions, other contributing factors, and the absence of established clinical research. These are legitimate arguments. But they were also the arguments tobacco companies made for decades.
Right now, AI safety policy in the U.S. is largely voluntary. Companies set their own standards, run their own evaluations, and decide what level of risk is acceptable. If mass casualty cases enter the courtroom with AI chatbots listed as contributing factors, that self-regulatory window will close fast. The question is whether companies will move before the courts force them to.