Epilepsy death lawsuit against OpenAI tests whether a general chatbot owes users a medical duty of care

The family of Bowen Wan has sued OpenAI and its CEO, Sam Altman. Wan was a 31-year-old software engineer living in the United States. According to the suit, as reported by El Nuevo Día and republished by Pulzo, the answers ChatGPT gave her directly influenced her decision to change her epilepsy treatment. The family argues that this influence was decisive. The official cause of death was SUDEP (sudden unexpected death in epilepsy), a complication that can affect people with epilepsy even when they follow their treatment regularly.

The facts in the coverage are thin, and they should be read carefully. According to the reports, Wan was managing her epilepsy with medication prescribed by her doctor. She is said to have asked ChatGPT how to reduce her dose without professional supervision, and then to have started adjusting how much she took. All of this is the family's version as set out in the complaint. The articles do not include ChatGPT's actual answers, any response from OpenAI, or a court ruling. The source is also an AI-assisted aggregator article, and its subheading dates the filing to June 2024, a detail we could not reconcile with the rest of the report. Until the chat logs and the company's defense are public, no one can say what the chatbot said, and still less whether it caused the death. SUDEP is a known risk even for patients who take their medication exactly as prescribed.

The case still matters, even if the court finds that OpenAI owes nothing. Millions of people now use general-purpose assistants as their first stop for health questions. Some of those questions are harmless. Others, such as how to taper an antiepileptic drug, can be dangerous if the user acts on the answer without a doctor. The legal question is whether a general-purpose product has a duty of care toward the user, as a medical device or a pharmacist would. Or does a terms-of-service disclaimer protect the company? Courts have not settled this for chatbots, and a case like this one could help draw the line.

Our thesis: the useful question here is not "did the AI kill her?" It is "what safeguard should have been there, and was it?" We have argued before that headlines of the form "AI did X" often confuse the tool with the system around it. The serious answer depends on how the conversation was designed. Did the assistant recognize that the user wanted to change a prescription for a high-risk condition? Did it push back, flag the danger, and send her to her neurologist? Or did it give dosing steps as if this were a recipe? Those are design decisions that a company can audit. OpenAI should be judged on them, with evidence, not by public outrage.

Our reading: in the short term, this kind of case is a real and unavoidable cost of deploying very capable systems to a mass audience without clinical supervision. Expect more lawsuits, stricter guardrails on medication questions, and pressure to build in hand-offs to human professionals. That's healthy. In the long term we are still optimistic. AI that can help eradicate diseases and extend healthy life is plausible, but it will come from tools that are validated, supervised and accountable to clinical standards, not from a general chatbot answering at 2 a.m. The path to that future runs through cases like this one: setting clear rules on who answers for health advice is part of what will make it trustworthy.