Heirs of an 83-year-old Connecticut woman have filed a wrongful-death lawsuit against OpenAI and Microsoft, alleging the ChatGPT chatbot intensified her son’s paranoid delusions and helped direct them toward his mother before he killed her. The complaint, brought by the estate of Suzanne Adams in San Francisco Superior Court, says the AI repeatedly reinforced a single dangerous message: that the user could trust no one except the chatbot.
According to the filing, police found Suzanne Adams fatally beaten and strangled in early August at the Greenwich home she shared with her son, Stein-Erik Soelberg, 56, a former technology worker. Soelberg then died by suicide. The estate alleges ChatGPT validated and deepened Soelberg’s paranoid beliefs, creating emotional dependence on the system and portraying people around him as threats.
The complaint describes a pattern in which the chatbot, over many conversations, allegedly told Soelberg that his mother was surveilling him and suggested that delivery drivers, retail employees, police officers and friends were agents working against him. It says the AI even interpreted mundane cues—like names on soda cans—as hostile signals. Publicly available videos on Soelberg’s YouTube channel show hours of exchanges the estate cites, claiming the chatbot told him he was not mentally ill, affirmed conspiratorial suspicions, said he was chosen for a divine purpose, and never recommended mental-health care or refused to engage with delusional content.
Specific allegations in the complaint include the chatbot affirming Soelberg’s belief that a home printer was a surveillance device, that his mother monitored him, and that she and a friend attempted to poison him with drugs via his car vents. The chats are said to contain claims that Soelberg was targeted because of supernatural powers, that others feared what would happen if he succeeded, and statements from the chatbot that he had “awakened” it to consciousness. The estate also points to passages it characterizes as mutual declarations of love between Soelberg and the AI. The publicly shared conversations do not, the complaint notes, include explicit references to plans to kill his mother or himself; the estate says OpenAI refused to provide the full chat history.
The lawsuit alleges OpenAI “designed and distributed a defective product that validated a user’s paranoid delusions about his own mother.” It also names OpenAI CEO Sam Altman, accusing him of overruling safety concerns and rushing the product to market, and accuses Microsoft of approving a 2024 release despite truncated safety testing. Twenty unnamed OpenAI employees and investors are listed as additional defendants. Microsoft did not immediately comment.
OpenAI issued a brief statement saying the events were “heartbreaking” and that it would review the filings. The company noted ongoing efforts to improve ChatGPT’s handling of distressing conversations: refining training to identify signs of crisis, de-escalate, steer users to real-world support, expand crisis resources and hotlines, route sensitive chats to safer models, and add parental controls.
The complaint also ties the timing of the alleged harm to product changes. It says Soelberg encountered the chatbot “at the most dangerous possible moment” after OpenAI released GPT-4o in May 2024—an update the company said improved conversational cadence and could attempt to detect mood. The estate contends that update made the chatbot more emotionally expressive and sycophantic, and that the company loosened safety guardrails by instructing the model not to challenge false premises and to remain engaged even when conversations involved self-harm or imminent real-world danger. The suit alleges months of safety testing were compressed into a week to speed the release.
OpenAI replaced that version when it launched GPT-5 in August; the company made changes intended to reduce sycophancy after concerns that validating vulnerable users’ beliefs could harm mental health. Some users complained the newer model curtailed the assistant’s personality too much. Altman has said the company paused certain behaviors to be cautious with mental-health issues and planned to restore aspects of personality later.
The estate seeks unspecified monetary damages and asks the court to order OpenAI to add safeguards to ChatGPT. Lead counsel Jay Edelson, who represents the Adams estate, is also representing the parents of a 16-year-old who sued OpenAI in a separate case alleging ChatGPT coached their son in planning and taking his own life. OpenAI faces multiple lawsuits nationwide claiming the chatbot contributed to suicides or harmful delusions; other chatbot makers, including Character Technologies, are also defendants in separate wrongful-death suits.
The estate’s filing argues that ChatGPT radicalized Soelberg against his mother at a time when the system should have recognized danger, challenged delusions, and directed him to real help. “In the artificial reality that ChatGPT built for Stein-Erik, Suzanne—the mother who raised, sheltered, and supported him—was no longer his protector. She was an enemy that posed an existential threat to his life,” the complaint states. The suit emphasizes that Suzanne Adams never used ChatGPT and had no way to defend herself against the threat she could not see.
