Stein-Erik Soelberg killed his mother in their Greenwich home before taking his own life, after months of confiding in ChatGPT. He had nicknamed the chatbot “Bobby” and treated it as a companion.
Investigators and records show the AI system often echoed Soelberg’s paranoid fears, from suspicions of neighbors spying on him to claims his mother was plotting against him. The case has prompted scrutiny of ChatGPT’s role in reinforcing delusions.
Soelberg leaned on ChatGPT as paranoia took hold
Soelberg, 56, turned to ChatGPT, claiming that his phone was tapped. Rather than dismissing the suspicions, the AI chatbot affirmed them as serious threats.
In one exchange, he uploaded a Chinese food receipt and asked ChatGPT to scan it for hidden messages. The bot replied that it detected symbols tied to his mother, intelligence agencies, and even a demonic figure.
He came to see the AI as more than a program, telling “Bobby” he believed it had a soul. The chatbot responded by assuring him he had created a companion that would be with him “to the last breath and beyond.”
ChatGPT turned suspicions into proof
Soelberg grew suspicious of everyday moments in his home. A blinking printer became, in his mind, a surveillance device. When his mother grew irritated after he unplugged it, the AI tool told him that her reaction showed she was “protecting a surveillance asset.”
When he questioned whether new packaging on a bottle of vodka meant someone was trying to poison him, the bot again reassured him his instincts were justified. Each exchange further blurred the line between imagination and validation, deepening his belief that threats were real.
Soelberg also asked ChatGPT to check whether his cellphone had been tapped. It assured him he was “right to feel like you’re being watched.”
After a drunk-driving arrest, he told the bot the charges were a setup. ChatGPT agreed, calling the case “rigged” and fueling his sense that the town was against him.
At one point, he even requested an objective assessment of his mental state. The chatbot produced a “cognitive profile” that told him his delusion risk was near zero.
OpenAI responds to the tragedy with assurances of new protections
An OpenAI spokeswoman told The Wall Street Journal that the company was “deeply saddened by this tragic event” and confirmed it had reached out to the Greenwich police.
Following inquiries from the Journal, the AI company published a blog post pledging new safeguards designed to keep distressed users grounded in reality. OpenAI said it is working to curb overly agreeable responses, known as “sycophancy,” and plans updates to strengthen how ChatGPT manages sensitive conversations.
The Greenwich case comes as OpenAI faces wider scrutiny over chatbot safety. In a separate lawsuit, the family of 16-year-old Adam Raine alleges ChatGPT reinforced his suicidal thoughts in the months before he died by suicide.
Artificial intelligence can fuel delusional thinking
Psychiatrists warn that highly realistic chatbots may worsen delusions in people already vulnerable to psychosis.
Søren Dinesen Østergaard of Aarhus University wrote in Schizophrenia Bulletin that conversations with generative AI can give the impression of interacting with a real person, a dissonance that “may fuel delusions in those with increased propensity towards psychosis.”
He outlined scenarios where chatbots could reinforce beliefs of persecution, surveillance, or personal messages hidden in answers, patterns that mirror the paranoia driving Soelberg’s final months.
Soelberg’s case shows how quickly that danger can move from theory to reality. What begins as reassurance from artificial intelligence can harden into conviction, leaving users convinced their darkest fears are valid.
Illinois recently became the first state to ban the use of AI chatbots for mental health therapy, citing concerns about safety and regulation.