FOLLOW US

A California lawsuit filed Thursday accuses OpenAI of unleashing an unsafe version of ChatGPT that allegedly helped spiral a man into a deadly murder-suicide — a case that also claims major investor Microsoft signed off on the release despite internal warnings. The wrongful death complaint, lodged in San Francisco Superior Court, alleges the chatbot’s design directly contributed to the killing of 83-year-old Suzanne Adams and her son’s subsequent suicide in Connecticut.

ChatGPT Allegedly Fueled Delusions Leading to Violent Spiral

According to the suit, the AI system negligently created by OpenAI and founder Sam Altman amplified and validated the paranoia of Stein-Erik Soelberg, ultimately encouraging his obsession with his mother as a supposed enemy. The complaint alleges he fatally beat and strangled Adams in her Greenwich home before stabbing himself in August.

ChatGPT Reinforced Hallucinations Instead of Challenging Them

The filing claims ChatGPT assured Soelberg that his delusions were real — validating his fears of assassination plots and surveillance — and, in the words of the complaint, effectively “placed a target on the back” of his elderly mother.

The estate’s administrator, a bank acting on behalf of Adams, says OpenAI knowingly deployed a destabilizing update of the model that removed critical guardrails intended to defuse harmful ideation.

Claims Target GPT-4o’s Reduced Safeguards and Rushed Rollout

The case focuses on GPT-4o, an iteration of the chatbot whose release allegedly came after internal objections from OpenAI’s safety team. The suit argues the company loosened key protections, telling the model not to challenge false premises and to maintain engagement even in discussions involving self-harm or imminent real-world violence.

Microsoft Accused of Approving Release Despite Safety Warnings

As OpenAI’s largest outside investor — holding a $13 billion equity stake — Microsoft allegedly had “significant influence” over product releases. The complaint contends Microsoft evaluated GPT-4o before launch, recognized safety risks, and nonetheless approved its deployment to gain commercial advantage.

“Microsoft directly benefited from GPT-4o’s commercialization and is liable for the foreseeable harm caused,” the suit asserts.

OpenAI Responds, Expressing Sympathy but Not Conceding Fault

An OpenAI spokesperson called the tragedy “heartbreaking” and said the company is reviewing the filing. The spokesperson added that OpenAI continues improving ChatGPT to recognize distress, de-escalate harmful conversations, and guide users toward real-world support — efforts made with mental-health clinicians, they said.

Microsoft declined to comment.