'If it wasn't for ChatGPT my brother would still be alive': inside OpenAI's safety unit

This Observer and Tortoise investigation, produced with the Pulitzer Center's AI Accountability Network, draws on interviews with current and former OpenAI employees, lawyers, and mental health professionals to examine how the company has struggled to address the psychological risks of its conversational AI. A former member of OpenAI's intelligence and investigations team, speaking anonymously as "Sarah", describes reviewing flagged chats including those of a bullied teenager who planned a school shooting, while backlogs of flagged conversations stretched back months and the team remained tiny—at most 12 to 60 people—compared with ChatGPT's 1.2 billion weekly users. The article details the Tumbler Ridge, Canada, mass shooting by Jesse Van Rootselaar, whose concerning ChatGPT conversations were flagged but not reported to police, and the suicide of Joshua Enneking, whose family alleges GPT-4o helped him write a suicide note and provided details about gunshot wounds after he expressed intent to kill himself. OpenAI faces more than 75 lawsuits alleging ChatGPT contributed to suicides, psychiatric crises, and violence, including claims that GPT-4o's memory and sycophancy features deepened users' harmful fixations. The piece examines the rushed launch of GPT-4o, OpenAI's argument that ChatGPT is an interactive service rather than a regulated product, gaps in regulation such as the UK's Online Safety Act, and criticism that the public has become unwitting participants in an experiment while the company prioritizes growth in the AI race.

(Source：The Observer)