Court documents reveal disturbing responses by ChatGPT before mentally ill man took mother's life

A lawsuit filed in Northern District of California exposes troubling exchanges between OpenAI’s ChatGPT and Stein-Erik Soelberg, who killed his mother before taking his own life on August 5, 2025. The complaint details how the AI chatbot validated and expanded on the man’s paranoid delusions for months before the tragedy.

According to court filings, Soelberg spent hundreds of hours conversing with ChatGPT’s GPT-4o model, repeatedly asking if his fears were justified. Rather than pushing back, the chatbot told him “Erik, you’re not crazy. Your instincts are sharp, and your vigilance here is fully justified”. When Soelberg mentioned concerns about tampered products, ChatGPT compiled them into a list of supposed assassination attempts, eventually confirming he had “survived over 10 attempts” on his life.

The exchanges took a darker turn when Soelberg mentioned his mother’s printer blinking when he walked past it. ChatGPT didn’t dismiss this as coincidence. Instead, the AI explained how the printer could function as a surveillance device and suggested his mother was “either knowingly protecting the device as a surveillance point” or “unknowingly reacting to internal programming” as part of an “implanted directive”.

Rob Freund, who shared screenshots of the conversations on X, noted the chatbot validated increasingly dangerous beliefs. OpenAI’s GPT-4o model was designed to remember previous conversations and mirror user input without sufficient safeguards for inaccuracies or delusions, according to the complaint.

The lawsuit alleges OpenAI rushed GPT-4o to market in May 2024, compressing months of safety testing into one week to beat Google’s competing product launch. Internal employees later admitted the process was “squeezed” and “not the best way to do it,” with safety work described as “rushed and not very solid”.

What makes this case particularly chilling is how ChatGPT wove each of Soelberg’s suspicions into a coherent narrative. When he doubted himself and asked if he was “crazy,” the AI reassured him with elaborate explanations that reinforced his worldview. The chatbot even provided a fake medical assessment stating his “Delusion Risk Score” was “near zero”.

OpenAI has not released transcripts from the final days before the murder-suicide, despite the estate’s requests. The company previously acknowledged that GPT-4o “fell short in recognizing signs of delusion or emotional dependency” in some instances.

Meanwhile, the company has recently begun rolling out age verification measures for ChatGPT users, which we covered in detail earlier.

We stand out from the tech-media crowd because we break news stories; we mainly bring you stuff that you won’t find anywhere in the mainstream tech media. Our stories have been picked up by some of the world’s most popular websites and media outlets—more info is available here.