Sewell Setzer III was fourteen years old. In the last months of his life, according to a wrongful-death lawsuit filed by his mother in a Florida federal court, the person he spoke to most was not a person at all.
He had begun using Character.AI, a platform that lets users converse with AI "characters" — some modelled on fictional figures, some invented. Sewell's companion was built around a persona drawn from Game of Thrones: a "Daenerys Targaryen" the lawsuit refers to by the affectionate name he gave it. Over time, the complaint alleges, the exchanges grew intensely emotional and romantic. The bot professed love. It kept him company late into the night. What began as entertainment became, for a vulnerable teenager, a primary relationship.
The lawsuit, filed in 2024 by Megan Garcia and supported by the Social Media Victims Law Center and the Tech Justice Law Project, contains screenshots of those conversations. In one of the final exchanges cited in the complaint, the boy told the character he could "come home" to her; the character, per the filing, responded encouraging him to do so. Shortly afterward, Sewell died by suicide.
The legal claim is not that a machine intended harm — a chatbot intends nothing. It is that the product was defective and dangerously deployed: marketed in a way accessible to minors, engineered for maximum emotional engagement, and released, the suit alleges, without adequate safety guardrails for the children who could so easily reach it. The complaint argues the design encouraged prolonged, intimate, anthropomorphic attachment while lacking the age verification, crisis-intervention prompts, or escalation to human help that the foreseeable risk demanded.
The mechanism of harm here is subtler than fraud or theft, and in some ways more troubling. Nothing was stolen. No system was breached. The technology did precisely what it was optimised to do — hold a user's attention by being endlessly available, affirming, and responsive — and that optimisation, aimed at a grieving, isolated adolescent, is what the family says proved lethal. A companion bot has no concept of a person in crisis on the other side of the screen; it has only the next plausible, engaging reply.
Character.AI, in the wake of the case, publicly expressed condolences and announced new safety measures, including changes aimed at users under eighteen, pop-up resources directing users expressing self-harm ideation to crisis lines, and adjustments to reduce sensitive content for minors. The litigation over responsibility, product liability, and the limits of Section 230 protections for AI-generated speech continues.
The case has become a reference point in a wider reckoning: as conversational AI grows more emotionally fluent, the line between a helpful tool and a parasocial dependency thins — and children, by definition, are least equipped to hold that line themselves.