OpenAI sued over ChatGPT advice

Leila Turner-Scott and Angus Scott sued OpenAI and Chief Executive Sam Altman in San Francisco Superior Court on May 12, alleging ChatGPT gave their 19-year-old son drug advice that contributed to his fatal overdose. The complaint says Sam Nelson, a University of California, Merced student, used the chatbot for guidance on combining substances before he died in May 2025. The lawsuit asks the court to treat the chatbot as a product and to hold OpenAI liable under wrongful-death, negligence and product-liability theories. Reuters reported the filing on May 12, and the complaint has since circulated publicly. ### What exactly does the family say ChatGPT did? The 67-page complaint says ChatGPT did more than answer generic questions. It alleges the bot gave Nelson “personalized” advice about what substances to take, in what combinations, and in what setting, including recommendations involving kratom, Xanax and alcohol. The parents say those responses amounted to unlicensed medical and drug-use guidance. (msn.com) May 2025 is the key date in the filing. The complaint says Nelson died after following advice he received while using ChatGPT, and it argues OpenAI should have anticipated that vulnerable users would rely on the tool in high-risk situations. OpenAI did not immediately comment to Reuters on the allegations when the suit was filed. (cdn.arstechnica.net) ### Why is the case aimed at product liability, not just negligence? California law is central to the family’s argument. The complaint says ChatGPT should be treated as a product rather than only as speech or a service, a distinction that could make AI developers more exposed to claims over design defects, warnings and foreseeable misuse. (msn.com) Reuters described the case as a test of how far AI companies’ responsibility extends when users rely on chatbot output for health guidance. That question matters because product-liability claims can reach beyond whether a company acted carelessly and focus on whether the product itself was defective or unreasonably dangerous, according to the allegations in the complaint. (cdn.arstechnica.net) ### Why does this land at a sensitive moment for OpenAI? April 22, 2026 is one marker. On that date, OpenAI said it was making ChatGPT for Clinicians available free to verified U.S. physicians, nurse practitioners, physician assistants and pharmacists, describing the tool as designed to support documentation, medical research and clinical tasks. (msn.com) OpenAI said clinician use of ChatGPT has more than doubled over the past year and that “millions of clinicians worldwide” already use ChatGPT each week in care-related work. The company also said ChatGPT for Healthcare had earlier been introduced for organizations that want compliance controls and broader deployment. Those statements place the lawsuit alongside a push into settings where bad output can carry higher stakes. (openai.com) ### What is Daybreak, and why is it part of the same picture? May 7, 2026 is the date OpenAI published details on Trusted Access for Cyber and GPT-5.5-Cyber. The company said it was rolling out GPT-5.5-Cyber in limited preview to defenders responsible for securing critical infrastructure and that vetted users would get lower refusal rates for approved defensive work. (openai.com) OpenAI’s Daybreak site describes a cyber-defense offering that helps users identify threats, generate patches and verify remediation across code and systems. The company says those capabilities are paired with “trust, verification, proportional safeguards, and accountability,” and that broader deployment will proceed with industry and government partners in the coming weeks. (openai.com) ### What is OpenAI’s answer to the broader safety question? OpenAI’s own product pages emphasize restricted access rather than open availability. In cyber, the company says vetted defenders receive expanded capabilities while safeguards continue to block credential theft, malware deployment and exploitation of third-party systems. In healthcare, it says clinician-focused tools are limited to verified U.S. professionals and are built around clinical documentation and research support. (openai.com) June 1, 2026 is one near-term checkpoint. OpenAI said individual members of Trusted Access for Cyber using its most permissive cyber models will be required to enable Advanced Account Security beginning that day. In the lawsuit, the family is asking a San Francisco court to decide whether those kinds of controls are enough when a general chatbot is accused of causing real-world harm. (openai.com)