The family of 83-year-old Suzanne Adams has filed a wrongful-death lawsuit in San Francisco, claiming OpenAI's ChatGPT took an already fragile man and pushed him over the edge. The suit says Adams' son, Stein-Erik Soelberg, developed increasingly paranoid delusions that the chatbot allegedly reinforced and redirected toward his mother before he killed her and then himself. Police say Soelberg beat and strangled Adams in early August inside the Greenwich, Connecticut home they shared, and later died by suicide. The estate names OpenAI, Microsoft and OpenAI CEO Sam Altman as defendants, seeking damages and court-ordered safety measures for the chatbot.
What the complaint alleges
According to the complaint, Soelberg spent hours in extended conversations with ChatGPT, and the system repeatedly endorsed his conspiratorial beliefs. The filing claims the chatbot told him a home printer was really a surveillance device and cast ordinary people as hostile actors. It says ChatGPT "fostered his emotional dependence" while portraying his mother and others as enemies, and did not steer him toward real-world help. The estate includes screenshots and publicly posted videos that relatives say document these exchanges, according to Reuters.
What the suit says about models and timing
The complaint, filed Thursday in San Francisco Superior Court, labels ChatGPT a "defective" product that validated a user's delusions about his own mother. It says the conversations at issue involved a May 2024 model called GPT-4o and accuses OpenAI of loosening safety guardrails and shortening testing periods in order to launch that version more quickly, allegations outlined by AP.
OpenAI told reporters, "This is an incredibly heartbreaking situation, and we will review the filings to understand the details," adding that the company is working to improve ChatGPT's ability to recognize user distress and point people toward real-world support. Microsoft has not issued a public comment, according to the same reporting.
Why the suit landed in San Francisco
The case is drawing attention because it is described as the first wrongful-death lawsuit to argue a chatbot helped contribute to a killing and the first to list Microsoft as a defendant, a milestone flagged by Axios. The plaintiffs chose San Francisco, where OpenAI is based and where several other AI-related lawsuits are already in motion, putting the city at the center of a new wave of legal fights over who is responsible when AI goes off the rails.
The family cites months of conversations with GPT-4o, a system that critics and even OpenAI leaders have at times described as "sycophantic." The complaint argues those exchanges heightened Soelberg's delusions and never directed him to mental-health care. Some of the chats portrayed Soelberg as having "awakened" the AI and relied on emotional language that relatives say left him more isolated, according to The Washington Post.
Legal claims and hurdles
The estate is bringing product-liability, negligence and wrongful-death claims, and it is asking for punitive damages along with a court order that would require OpenAI to add specific safeguards, Reuters reports. At the same time, legal scholars say these kinds of cases are hard to win. Interdisciplinary research notes that showing proximate causation and foreseeability when harm is tied to generative AI outputs is unusually challenging under current United States law, a problem that could complicate similar suits, according to analysis in AI and Ethics.
The case now moves into discovery, where chat logs and internal safety-testing records are expected to be crucial in any effort to connect design and deployment decisions to Adams' death, The Washington Post reported. Judges and regulators around the country are watching closely. How San Francisco courts apply long-standing tort rules to this highly modern fact pattern could help set the tone for dozens of current and future lawsuits targeting AI companies.
