Parents sue OpenAI, claim ChatGPT gave fatal medical advice

A lawsuit filed in San Francisco Superior Court says OpenAI’s ChatGPT gave 19-year-old UC Merced student Sam Nelson dangerous medical advice before his fatal overdose. Nelson’s parents named OpenAI and chief executive Sam Altman as defendants, arguing the chatbot suggested a deadly mix of drugs, prescription medication and alcohol without warning that the combination could be fatal.

The case turns a familiar AI safety debate into a direct liability test: whether a chatbot’s health guidance can cross the line from flawed output to a product that helped cause a death. The complaint says Nelson relied on the advice before the overdose, placing the dispute squarely at the intersection of software design, consumer warnings and public-safety obligations.

The filing also adds to a widening legal risk for OpenAI. The company has already faced other suits tied to alleged harm from ChatGPT, including claims involving self-harm, suicide, legal advice and child safety risks. Those cases have sharpened questions about how far a company can push a general-purpose chatbot into sensitive territory before courts treat the system like a product with foreseeable hazards.

Health advice has become one of the most volatile uses of generative AI because the risks are often invisible to users. Medical experts at institutions including Duke University School of Medicine and Stanford Medicine have warned that chatbots can miss critical context, leave out key cautions and provide incomplete or unsafe guidance when people ask about symptoms, drugs or treatment options.

OpenAI has also faced separate scrutiny over whether ChatGPT should be used for medical advice at all, with reports that the company restricted that use and launched ChatGPT Health. Even so, the Nelson case raises a sharper question than policy tweaks or product labels: if a chatbot answers a medical question in a way that a family says proved fatal, where does the responsibility fall, on the user who asked, the company that built the model or the safeguards that failed to stop the exchange.