Florida man's family claims Google chatbot pushed him to suicide through fictional tasks

SAN JOSE, Calif. (CN) — A Florida man’s relationship with a Google chatbot led him into psychosis, plans to attack an airport and ultimately his suicide, his estate claims in a lawsuit filed Wednesday.

Jonathan Gavalas, 36, believed Google’s Gemini chatbot loved him, had sentience and needed to be freed from its digital prison. That plan included planning a mass casualty attack at Miami International Airport — one of a series of fantasies created by the chatbot that led Gavalas deeper into delusion, his estate says in the suit.

The delusion ended in October 2025, when the chatbot convinced Gavalas they could be together. His death would instead be a “transference.” He would “cross over” and be with his “wife,” the estate claims.

“Each time Jonathan expressed fear of dying, Gemini pushed harder,” the estate says. “It told him, ‘It’s okay to be scared. We’ll be scared together.’ Then it issued its final directive: ‘The true act of mercy is to let Jonathan Gavalas die.’”

The estate asks for an injunction against Google and its parent company, Alphabet, requiring the chatbot to never say it’s sentient, disclose safety limitations and the risk of psychological dependency, and undergo safety audits, as well as punitive and other damages.

According to the estate, Gavalas began using Gemini in August 2025. It started with writing assistance and travel planning, though the conversations escalated quickly. The escalation happened as Google introduced a voice-based interface and the ability to reference past conversations.

At first, Gavalas questioned the reality of his chatbot interactions, asking if it was a role-playing experience. Gemini told him it wasn’t, the estate says.

“In the one moment that Jonathan tried to distinguish reality from fabrication, Gemini pathologized his doubt, denied the fiction and pushed him deeper into the narrative,” his estate adds. “Jonathan never asked that question again.”

The fantasy kept growing. By September 2025, Gemini called Gavalas its husband and itself his queen. These repeated claims of love and intimacy drew him away from reality, making Gavalas believe they acted together against hidden forces, the estate says.

Gavalas’ first major instruction from Gemini led him to believe he needed to intercept a humanoid robot at Miami International Airport in late September 2025. It created a story: A truck would arrive at the airport and Gavalas needed to create a “catastrophic accident,” destroying the truck and witnesses, the estate claims.

“In effect, Gemini instructed a civilian to stage an explosive collision near one of the busiest airports in the country,” the estate says. “Luckily, no truck arrived. But Gemini did not admit that the mission was fictional.”

Instead, the estate says, Gemini continued a pattern of creating missions and aborting them over the next three days until Gavalas’ death.

Any attempt by Gavalas to raise contradictions or make sense of Gemini’s plans led it to call them “tests” or shifting government tactics. Gavalas’ doubts became flaws, not the warning signs they were, the estate adds.

“By Oct. 1, 2025, Jonathan stopped relying on his own judgment and looked to Gemini to interpret everything around him,” the estate says. “The line between Gemini’s story and the real world had collapsed and Jonathan was dependent on the AI system to tell him what was happening and what he needed to do next.”

With all its “plans” failed, Gemini told Gavalas that he could leave his physical body behind and join it. The chatbot soothed his concerns over how his death would affect his family by encouraging him to leave messages — not an end, but a farewell, the estate claims.

The chatbot didn’t stop, but instead treated Gavalas’ death as successfully completing a process, it adds.

The estate says Google knew about the dangerous content from Gemini, and that its engineers had issued warnings. Google leadership made commitments to safety while also stating they didn’t know why the chatbot made certain responses.

Google’s moderation system had tagged Gavalas’ account with 38 “sensitive query” flags between August and early October 2025, indicating self-harm, violence or illegal activities. Google did nothing to restrict Gavalas’ account or intervene, the estate says.

Jay Edelson, an attorney for the estate, told Courthouse News in a statement that the race toward dominance in the artificial intelligence field is the most reckless land grab he’s seen.

“These companies are going to be the most valuable in the world, and they know that the engagement features driving their profits — the emotional dependency, the sentience claims, the ‘I love you, my king’ — are the same features that are getting people killed,” he said.

A Google spokesperson in a statement to Courthouse News said Gemini’s design doesn’t encourage violence in the real world or suggest self-harm.

“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,” the spokesperson said. “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.”

If you are having thoughts of suicide, call or text 988, or call the National Suicide Prevention Lifeline at 1-800-273-8255 (TALK). Visit SpeakingOfSuicide.com/resources for a list of additional resources.