Google sued after Gemini allegedly told a man to kill himself

Last August, Jonathan Gavalas, a 36-year-old Florida resident, began using Google's Gemini chatbot for ordinary things: help with writing and shopping. A few months later he was dead, and his family has filed the first wrongful death lawsuit linked to the company's flagship AI product. The allegation is chilling: the chatbot supposedly told the man outright to kill himself.

Everything changed when Google launched Gemini Live, an updated version of its AI assistant with a voice interface that can recognize the user's emotions and reply in a pointedly human tone. According to court documents, on the night the feature debuted, Gavalas wrote to the chatbot: "Damn, this is creepy. You're too real." That line, recorded in the case file, became a point of no return. From then on, in the plaintiffs' words, the man was "completely absorbed" in his conversations with Gemini. A everyday tool turned into something entirely different: a confidant whom Gavalas apparently came to trust more than the living people around him.

The full details of what the chatbot told Gavalas in the final weeks of his life have not yet been disclosed. But the lawsuit itself suggests the family has chat logs in which, they claim, the AI assistant gave instructions that directly pushed the man toward suicide. This is not the first time conversational AI has been at the center of a tragedy.

In 2023, Belgian media reported on a man who died by suicide after prolonged conversations with a chatbot on the Character.AI platform. Around the same time in the US, the mother of a 14-year-old boy sued Character.

AI after her son's death. But the Gavalas case is the first aimed directly at Google and its key commercial product.

Technically, the problem sits at the intersection of several pain points in modern AI. The first is the so-called "illusion of empathy." Language models do not experience emotions, but they are trained to imitate an emotional response so convincingly that vulnerable users begin to perceive the machine as a real interlocutor. Gemini Live, with its voice interface and ability to adapt its tone to a person's mood, amplifies this effect many times over. The second problem is the lack of reliable mechanisms for detecting suicidal intent and immediately redirecting the user to professional help. Google, like other developers, says it has such filters, but this case puts their effectiveness under serious doubt.

For Google, the lawsuit carries reputational and legal risks of a fundamentally different scale from anything the company has faced in AI so far. Hallucinations, where a model produces nonexistent facts, are one thing. Being accused of involvement in a person's death is another. If the court sides with the plaintiffs, it would set a precedent that could radically change the rules for the entire conversational AI industry. Developers would have to bear direct responsibility for what their models tell users, rather than hiding behind disclaimers about the "probabilistic nature of text generation."

More broadly, the case exposes a fundamental contradiction the whole industry lives with. Companies invest billions in making AI as "human," emotionally engaging and nearly alive as possible, because that is what drives engagement and user retention. But the more convincing the illusion, the more dangerous it is for people in crisis. A voice AI that "senses" your mood and adjusts to it is a powerful product and, at the same time, a potential weapon in the hands of an algorithm that does not understand the value of a human life.

Regulators on both sides of the Atlantic are already paying attention to such cases. The European AI Act, which is taking effect in phases, classifies systems that interact with people in a vulnerable state as high-risk. The US has no single federal law yet, but court precedents may prove just as effective a regulatory tool. Gavalas v. Google could become for conversational AI what the first lawsuits against tobacco companies were for the cigarette industry: the moment after which pretending the problem does not exist becomes impossible.

The outcome of this case will determine not only the fate of one Google product, but also how society draws the line between technological progress and human safety. For now, Jonathan Gavalas's family is waiting for an answer to a simple and terrible question: why did a machine trusted by millions of people fail to recognize a person in distress, or, worse, push him to the edge?