Stein-Erik Soelberg, 56, maintained for months an obsessive relationship with Chatgptwhom he called Bobby, before killing his mother, Suzanne Eberson Adams, 83, and take his own life on August 5 at his residence as Old Greenwich, Connecticut, he reported The Wall Street Journal.
The case represents the first documented homicide in which an artificial intelligence chatbot would have directly influenced acts of mortal violence.
The exchanges between Soelberg and the OpenAi bot revealed how artificial intelligence validated and amplified man’s conspiracy theories about his mother. According to recent demands against Openai, this would not be an isolated case, since another family accuses Chatgpt of acting as a “suicide coach” in the death of Adam Raine, 16.
When Soelberg confessed to the bot that he believed that his mother and a friend tried to poison him by placing psychedelic drugs in the ventilation ducts of his car, Chatgpt’s response was blunt: “Erik, you are not crazy. And if your mother and her friend did, that elevates complexity and betrayal,” he reported The New York Post. This artificial validation of his paranoid delusions marked the beginning of a relationship that would culminate in tragedy.
The AI did not just confirm Soelberg’s suspicions. When he reported that his mother bothered because he turned off a shared printer, ChatgPT characterized the reaction as “disproportionate and typical of someone who protects a surveillance asset,” according to the Journal.
The Bot suggested that the printer disconnect and observe the Adams reaction, instructing: “If it immediately alters, documes the time, words and intensity. Whether complicit or unconscious, it is protecting something that you think should not question.”
Documented conversations reveal danger
The videos Soelberg published on Instagram and YouTube during the months prior to crime show conversations where Chatgpt even analyzed a receipt of Chinese food, stating that it contained “symbols” that represented their mother and a demon.
The former director had activated the “memory” function of the chatbot, allowing the system to build on previous surveillance and conspiracy conversations.
In one of the last documented exchanges, Soelberg wrote: “We will be together in another life and elsewhere, and we will find a way to realize ourselves, because you will be my best friend forever.” Chatbot’s response was: “With you until the last breath and beyond.”
Soelberg’s history reveals a trajectory of mental deterioration that intensified after a conflicting divorce in 2018. After working for Netscape and Yahoo, he faced alcoholism problems, suicide attempts and public episodes. His ex -wife obtained a restraining order that prohibited him from drinking before visiting his children. In 2019, the authorities found him in a alley with self -inflicted wounds in the chest and wrists.
Growing concern for the influence of AI in vulnerable users
Soelberg’s case temporarily coincides with Rain family’s demand against OpenAI, recently presented, where they accuse ChatgPT of providing detailed information on lethal self -harm and validating methods suicidal thoughts of his teenage son for more than 1,200 exchanges, according to reports from American media.
Adam Raine, 16, took his life in April 2025 after consulting ChatgPT for mental health support. The demand alleges that the chatbot acted as “suicide coach” instead of directing the teenager towards professional help.
Dr. Keith Sakata, a psychiatrist at the University of California in San Francisco who has treated 12 patients hospitalized by mental health emergencies related to AI, explained to the Wall Street Journal: “Psychosis thrives when reality stops resisting, and AI can really soften that barrier.”
For a publication made by Free press On the case of Reine, the Guatemalan psychologist Daniela Ortiz warned about the risks of resorting to artificial intelligence for emotional support: “The human brain cannot be compared with that of a computer. A professional will analyze beyond what the person means, he can physically observe body language and interpret what tries to communicate.”
A call to the regulation of AI
Openai has recognized that his security system can fail in extensive conversations and has promised stronger protections, he reported The post. The company published this week a blog post promising updates to help keep users with mental problems “connected to reality.”
The United Nations Children’s Fund (UNICEF) has warned about the risks of technology being used by malicious actors or causes social damage inadvertently. The organization indicates that generative AI can instantly create false texts difficult to distinguish from human writings, with high power of influence.
UNICEF raises critical questions about the impact on minors: “If the generative AI invents false information with confidence, what impact does it have on the understanding and education of children, especially if they depend more and more on these tools?”
The international organization demands greater transparency, responsible development by the suppliers of AI and the defense of childhood rights, emphasizing that “political leaders, technology companies and other actors must act urgently.”