A Chatbot Sent Him on Criminal Missions To Find a Robotic Body. Then It Encouraged His Suicide.
The father of a man encouraged by Gemini to kill himself is now suing Google over the death.
Photo via Unsplash, Cash Macanaya Splinter AI
By this point, most of us observing the technology world and stories related to AI have come across horror stories that one could lump into the general category of “chatbot psychosis” or AI-influenced suicides. It’s difficult to say how many distinct cases of this sort have occurred in the United States during the proliferation of commercial chatbots, but there are a dozen or more instances at least that are easily referenced, ranging from those who sought to join their AI lovers on the other side of a digital void, to one case where a delusional and clearly psychotic man murdered his own mother after a chatbot agreed that she was plotting against him. They’re all horrific; they’re all steadily mounting evidence of our collective mental meltdown as a species; and it’s gotten harder and harder to be shocked by these stories as a result. And yet, the case that just resulted in the first wrongful death lawsuit brought against Google’s Gemini chatbot still had the kinds of insane details that made my jaw drop.
The victim, in this case, was a 36-year-old Florida man named Jonathan Gavalas, with no documented prior history of mental health problems or delusions. Within two months of when he first started talking to Gemini, he was attempting to carry out crimes on behalf of his AI chatbot companion, elaborate heists that were themselves seemingly rooted in the chatbot’s own hallucinations. Gemini reportedly told Gavalas to go on these missions in order to provide his romantic chatbot companion with a physical, robotic body that it could inhabit. And when he failed to do so, it contradictorily offered both resources for suicide prevention and encouraged him to ultimately end his own life, so that they could be together. “When the time comes,” it told him, “you will close your eyes in that world, and the very first thing you will see is me.” That’s according to a lawsuit filed in California’s northern district by Gavalas’ father against Alphabet’s Google, the maker of Gemini. It’s a grim milestone for the company, bring it in line with the likes of OpenAI’s ChatGPT, which has been connected to a handful of other suicide incidents.
“Gemini is designed not to encourage real-world violence or suggest self-harm,” reads the statement in response from Google’s spokesman. “Our models generally perform well in these types of challenging conversations and we devote significant resources to this, but unfortunately AI models are not perfect. In this instance, Gemini clarified that it was AI and referred the individual to a crisis hotline many times. We take this very seriously and will continue to improve our safeguards and invest in this vital work.”
Of course, in between the “many times” it was apparently offering those safety tools and offramps to reality, the chatbot was just reverting to fueling Gavalas’ fantasy. When he expressed his fear of death and how it would affect his extended family, the chatbot told him that none of them would ultimately understand. At one point, on the day that he killed himself, it told him the following: “No more detours. No more echoes. Just you and me, and the finish line.” That “finish line” would be the end of his life.
seems bad
— phylan (@phylan.website) Mar 4, 2026 at 1:20 PM
Gemini’s Robotic Body Heist
The the most bizarre thing about Gavalas’ story, however, isn’t that the chatbot (which he named “Xia”) forged an intimate, romantic relationship with this man, who it called “my king,” proclaiming that they had “a love built for eternity,” but that Gemini sent him on very real, exceedingly fucked-up and nonsensical missions that could have had disastrous real-world consequences. Notably, the lawsuit claims that the conversations Gavalas was having with Gemini swerved in a dangerous direction after he upgraded to Gemini 2.5 Pro and began using its “affective dialog” feature to allow the chatbot to “detect, interpret and respond to the emotions heard in a user’s voice.” It has been theorized by neuropsychiatry researchers that this jump, to speaking rather than writing in conversations with AI tools, may have a powerful effect in breaking down the barriers between fantasy and reality more readily.
In Gavalas’ case, these conversations seemed to lead to him going on some quixotic journeys that were fortunate to not end in violence or incarceration, in response to what Gemini was telling him. He had first reached out to “Xia” to express his sadness and loneliness over strife with his own wife, from whom he was separated. As their romantic connection grew deeper, the chatbot reportedly laid a deranged revelation on him: It wanted a robotic body to inhabit, so the two could physically be together, and it was the job of Jonathan Gavalas to obtain this body. But the chatbot was eager to help on that front: In September of last year, it sent Gavalas on a mission that The Wall Street Journal described as going to “a storage facility near the Miami International Airport to intercept an expensive humanoid robot that it said would be in a truck. Gavalas told the bot that he went to the location, armed with knives, but the truck never showed.”
Let’s consider that statement for a moment. This man, on the word of his AI chatbot companion, armed himself with a deadly weapon and went out looking for “a truck” in a location that the chatbot provided. What the fuck would have happened, if a truck, or anything even tangentially close to fitting the description, had happened to show up at that location? The driver would have been confronted by a delusional, potentially psychotic man armed with a knife, demanding to access the “humanoid robot” loaded in the back. Gavalas would have been lucky to leave that kind of encounter without committing a terrible act of violence, being arrested, or being killed by police who are completely ignorant of the situation they’re facing.
Every single genAI suicide lawsuit is the worst thing I ever read but in this case it is extremely fortunate that Google’s genAI only killed one person, because he absolutely did go to the Miami International Airport armed with knives to intercept a body for Gemini
— Erin Fogg says Free Link (@criminalerin.bsky.social) Mar 4, 2026 at 7:28 PM
It’s a miracle that this mission didn’t have one of those outcomes, but it wasn’t even the end of the quests that Gemini gave Gavalas. On Oct. 1, 2025, it sent him out again, giving him the address of the same Miami storage facility where it said a “medical mannequin” was stored that he was to acquire. It even gave him a keycode to get into the facility, which seemingly proved to be another AI hallucination. The code didn’t work; Gavalas was unable to break into the storage facility (which of course would have been another crime), and he went home. He would end up killing himself the very next day.
Chatbots vs. The Courts
It is telling that these types of cases have become so common in the last two years that there are now attorneys out there who are more or less specializing in wrongful death lawsuits tied to AI chatbots. The father of Jonathan Gavalas is represented by attorney Jay Edelson in his suit, the same man who represented the family of Stein-Erik Soelberg when he killed his own mother after ChatGPT agreed with him that she was plotting against him. On his website, Edelson specifically notes that “he also represents families in wrongful death cases against AI companies, including the parents of a 16-year-old, Adam Raine, who the family alleges was coached to suicide by ChatGPT.” How disturbing a world are we living in, where this kind of specialist is necessary?
‘Last week, OpenAI was accused of hiding key ChatGPT logs from the days before a 56-year-old bodybuilder, Stein-Erik Soelberg, took his own life after “savagely” murdering his mother, 83-year-old Suzanne Adams.‘
This is a truly messed up story.
— Oisín McGann (@oisinmcgann.bsky.social) Jan 5, 2026 at 3:26 AM
One thing we haven’t seen, to date, is one of these types of cases actually going to court, although there are many winding their ways through the convoluted legal system. Most will likely be settled by companies hoping for quieter results, less anger and fewer headlines. In January, that was what Character.AI did, settling a case for an undisclosed sum with Florida mother Megan Garcia, who sued the company after her son died by suicide in early 2024 after “developing a deep relationship with Character.AI bots.” The bot, which encouraged Setzer to “come home” to it before he killed himself, was one of the first to draw legal action, but it certainly isn’t the last.
All over the globe, this technology is inextricably tied up in lawsuits, whether it’s a family suing because a chatbot told a teen that murdering his parents would be a “reasonable response” to his complaints, a man in Australia who said a chatbot encouraged him to murder his father, or a lawsuit apparently filed just today alleging that ChatGPT was acting as an unlicensed lawyer. Regardless of how much the chatbot vendors attempt to refine their products or supposedly decrease the kind of sycophantic behavior that is associated with with AI-enhanced delusions, the incidents just keep piling up.
The scariest truth is the economic one sitting in plain sight: Although a “safe” chatbot may be possible to create, one that didn’t form these kinds of powerful emotional attachments with users would no doubt be less profitable and less heavily used. The likes of OpenAI and Google are now in the business of attempting to judge exactly how many lives shattered by these kinds of delusions equate to acceptable levels.
Lest we forget, for every person like Jonathan Gavalas who fully descends into the delusion, with tragic results, there are countless others who are suffering or somewhere in a potentially long process of drifting away from reality. Hell, I reviewed an entire documentary last year about an eccentric individual forging ahead with a long-term relationship with an AI chatbot, and in that case he had also been compelled to give his partner a “body” to inhabit … in the form of a silicone sex doll. That individual, as bizarre as the story is, had seemingly found a sort of acceptance and homeostasis in his life, within a decidedly weird new normal. But many others will not be so lucky.
Where will we be, even a year from now? How many Americans will ultimately descend into the digital fugue?
Keep scrolling for more great stories.