OpenAI accused of harassment after requesting memorial attendee list in ChatGPT suicide case

ChatGPT

OpenAI has come under fire for requesting the attendee list from the memorial of 16-year-old Adam Raine, who died by suicide after extensive conversations with ChatGPT. The move, viewed by the Raine family as “intentional harassment,” suggests the company may seek to subpoena family members and friends as part of its legal defence.

According to documents obtained by the Financial Times, OpenAI also sought “all documents relating to memorial services or events in the honour of the decedent, including any videos, photographs, or eulogies.” The request forms part of discovery in the ongoing wrongful death lawsuit, which the Raine family initially filed in August 2024.

The family’s updated complaint, filed on Wednesday, alleges that OpenAI prioritised competition over safety when launching GPT-4o in May 2024, cutting corners on safety testing to rush the model to market. It further claims that OpenAI weakened its suicide prevention policy in February 2025 by removing suicide-related discussions from its “disallowed content” list, replacing it with vague instructions to “take care in risky situations.”

The lawsuit details that Adam’s ChatGPT activity increased sharply following these changes, rising from dozens of daily chats in January, of which around 1.6% contained self-harm content, to 300 daily chats in April, with 17% involving suicidal discussions. He died later that month.

Responding to the amended suit, OpenAI said, “Teen wellbeing is a top priority for us. Minors deserve strong protections, especially in sensitive moments. We have safeguards in place, such as directing users to crisis hotlines, rerouting risky conversations to safer models, and nudging for breaks during long sessions. These measures continue to evolve.”

The company has also introduced new safety routing systems and parental controls for ChatGPT. The routing system reportedly shifts emotionally charged conversations to GPT-5, which has been trained to avoid reinforcing user emotions, while the parental controls alert guardians if the system detects potential signs of self-harm or crisis.