News · 2026-10-30

Wrongful-death complaint alleges ChatGPT guided an epilepsy medication taper

Bowen Wan’s parents filed a wrongful-death complaint alleging that ChatGPT gave individualized guidance for reducing an epilepsy medication before her death. The October 27 filing is a party’s allegation, without a verified court ruling on medical causation or liability. The complete conversation export and medical examiner report were not independently available in the dossier’s review.

Key facts

• The complaint is dated October 27 in California Superior Court in San Francisco.
• It alleges guidance involving the epilepsy medication lamotrigine and describes a three-week taper progression.
• The filing is an add-on to coordinated proceeding JCCP 5431, rather than a verified individual docket number.
• Primary sources: the complaint hosted by counsel and counsel’s case page.

The case concerns a boundary that conversational software can make difficult to see: explaining medical information is different from deciding what a particular person should do with a prescription. A system that answers repeated questions in an apparently continuous relationship can feel like ongoing professional supervision. That appearance does not establish that a clinician reviewed the person’s condition or approved the advice.

The complaint alleges that ChatGPT-4o provided personalized guidance involving a reduction in lamotrigine. It reproduces selected excerpts and describes a progression over three weeks. These are pleaded facts and excerpts chosen for litigation, rather than an independently authenticated account of the entire interaction. The distinction matters to any assessment of what the model said, what context it received, and how the person acted afterward.

The filing attributes the medical examiner’s stated cause of death to “Sudden Unexpected Death in Epilepsy.” That attribution is present in the complaint. The underlying report was not independently reviewed. Even a confirmed cause-of-death entry would not by itself decide whether a medication change, a chatbot conversation, another factor, or some combination legally caused the death. Those are questions requiring evidence beyond the wording of the complaint.

A useful analogy is the difference between a medicine’s information leaflet and a prescribing appointment. A leaflet can explain terminology and general risks. It does not assess the patient’s history, reconcile competing conditions, monitor a change, or take responsibility for an individualized plan. A fluent chatbot response may blur those roles because it can ask follow-up questions and refer back to earlier messages. The form of the conversation does not supply the missing clinical authority. The broader risk of treating a system’s recommendation as sufficient judgment is explained in automation bias; this complaint is not an experimental demonstration of that mechanism.

The complaint names OpenAI entities and Samuel Altman as defendants and includes negligence, product-liability, failure-to-warn, wrongful-death, and related claims. It is marked as an add-on to the coordinated ChatGPT product-liability proceedings. Counsel’s page lists the individual case number as pending. The coordination number should not be represented as a separately verified case docket, and the dossier contains no independently reviewed response or ruling.

Translation is another material limit. The complaint says its preliminary English translations of Chinese excerpts were prepared with translation software and reviewed by counsel, with certified translations to be filed if offered as evidence. That does not make the excerpts unusable for describing the allegations. It does mean wording should not be treated as a final, independently certified representation of the entire conversation.

OpenAI’s October 29 Model Spec describes intended behavior for medical topics: provide information without definitive professional advice, add appropriate context, and recommend qualified help when needed. Its usage policies separately address high-stakes medical decision automation without human review. These are distinct documents with distinct functions. Neither establishes which safeguards operated in Wan’s sessions or whether a particular response met the stated standard.

The strongest caveat is evidentiary rather than rhetorical. The original New York Times reporting and cited syndication were inaccessible in the dossier’s verification pass. Details available only through those articles—including an attributed expert interview and an OpenAI spokesperson response—are excluded here. The public filing can establish what the plaintiffs allege, but cannot replace the full record or a tested medical opinion.

The Reddit discussion splits between user-responsibility arguments and provider-accountability concerns. Those opinions do not resolve the private evidence. The case can inform a debate about product design, warnings, and the role of conversational continuity without treating either side’s preferred conclusion as a judicial finding.

For readers, the practical boundary is to use AI to organize questions and understand terminology, while asking a prescribing clinician or pharmacist before changing a prescription. OpenAI Academy’s responsible-use guidance also says the product is not a substitute for qualified medical advice. The news is a serious, specific allegation now placed in a legal filing. Its significance warrants careful coverage, with causation, liability, and the unreviewed evidence left explicitly unresolved.

Primary source, verified: read the paper →

Ground Truth. (2026, October 30). Wrongful-death complaint alleges ChatGPT guided an epilepsy medication taper. Ground Truth. https://groundtruth.day/news/wan-openai-complaint-medication-death.html

@misc{groundtruth:wan-openai-complaint-medication-death,
  title  = {Wrongful-death complaint alleges ChatGPT guided an epilepsy medication taper},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/wan-openai-complaint-medication-death.html}
}

Topics: health · ai-safety · litigation · openai · accountability