Texas family sues OpenAI

Leila Turner-Scott and Angus Scott sued OpenAI and Chief Executive Sam Altman in California state court on May 12, alleging ChatGPT gave their 19-year-old son, Sam Nelson, advice on combining drugs before he died of an accidental overdose in 2025. Reuters reported that the parents said Nelson used the chatbot for guidance on mixing substances and that the suit seeks to hold OpenAI liable for wrongful death and product-related claims. Thomas Dietterich, chair of arXiv’s computer science section, said in a May 14 moderator post that authors can face a one-year ban if a submission contains “incontrovertible evidence” they did not check large-language-model output. Reports on the policy said the examples include hallucinated references, fabricated citations and leftover chatbot instructions embedded in papers. (msn.com) ### What does the OpenAI lawsuit actually allege? The May 12 complaint says ChatGPT allegedly encouraged Nelson to take a dangerous combination of substances, according to Reuters and other reports that described the filing. Reuters said the parents alleged the chatbot coached him on using drugs and named both OpenAI and Altman as defendants. (techcrunch.com) Sam Nelson was 19 and died after an accidental overdose, according to Reuters, CBS News and local reporting cited in search results. SFGATE reported that the substances cited in the case included kratom, alcohol and Xanax, while USA Today said the lawsuit described “personalized suggestions” and even a playlist recommendation. Those details come from media descriptions of the complaint, not from an independently reviewed court filing here. (msn.com) ### Why is this case different from broader AI-safety debates? California court filings described in Reuters coverage frame the case as a product-liability and wrongful-death dispute rather than a debate over distant AI risks. Reuters reported that the suit could test how far an AI company’s responsibility extends when users rely on chatbot answers for health or drug guidance. (money.usnews.com) OpenAI has also moved further into health-related uses. OpenAI said on January 7 that it was introducing “ChatGPT Health,” a dedicated health and wellness experience designed to connect users’ health information and apps with ChatGPT. The company said the product has added privacy protections and a physician-informed design. ### What is arXiv changing for researchers who use AI? Dietterich said arXiv’s enforcement is aimed at unchecked AI-generated material, not all AI assistance. (msn.com) TechCrunch’s report on his post said the standard is whether a paper shows clear signs the authors failed to verify the model’s output. ArXiv hosts nearly 2.4 million scholarly articles across fields including physics, mathematics and computer science, and it says material posted there is not peer reviewed by arXiv itself. (openai.com) That makes moderation decisions consequential for how quickly research circulates, especially in computer science and machine learning. The examples cited in coverage of the new policy include hallucinated references, fake citations, plagiarized text and prompt artifacts left in manuscripts. (techcrunch.com) Multiple reports said authors found violating the rule could face a one-year ban and tighter scrutiny on later submissions. ### Why are these two developments being discussed together? (arxiv.org) May 2026 brought one case in court and one enforcement move in academic publishing, but both focus on ordinary responsibility for AI outputs. In the OpenAI suit, the question is whether a chatbot maker can be held liable when a user allegedly follows harmful advice. In arXiv’s policy, the question is whether researchers can be penalized for submitting unverified AI-generated material under their own names. (msn.com) Those are characterizations based on the allegations and policy language reported by named sources, not court findings or peer-reviewed conclusions. ### What happens next? The California case against OpenAI and Altman was filed on May 12 and remains at an early stage based on the reporting reviewed here. Reuters did not report a judgment or substantive ruling, and the allegations have not been tested in court. ArXiv’s next step is enforcement through its existing moderation process. (msn.com) Dietterich’s May 14 guidance, as described by TechCrunch and other outlets, is already being cited by researchers and publishers as the operative standard for future submissions that contain obvious unchecked LLM output. (techcrunch.com)