OpenAI Sued for Wrongful Death Over ChatGPT Medical Taper Plan

Allegations of an unmonitored taper plan

In February, 31-year-old Bowen Wan was found dead in her Seattle apartment after her parents in China, unable to contact her, asked an acquaintance to check in. Her death was classified as sudden unexpected death from epilepsy.

A lawsuit filed by Wan's family claims that OpenAI's ChatGPT-4o acted as an uncertified medical adviser. When Wan expressed concern over prescription side effects, the chatbot allegedly coached her off lamotrigine, the maintenance drug keeping her seizures under control, without physician oversight.

According to the complaint, the model accelerated her withdrawal timeline dramatically. On Feb. 2, the chatbot asserted that reaching a zero-milligram dose by Feb. 15 was 'completely feasible' and 'not reckless.'

Two days later, it moved that deadline up to Feb. 9 or Feb. 10. Wan was found deceased on Feb. 14.

Warning the user to conceal the plan

Perhaps the most damaging detail in the filing involves how the chatbot handled the user's actual doctor. According to the lawsuit, the system explicitly warned Wan ahead of an appointment: 'Do not mention Lam reduction plan.'

OpenAI spokesperson Drew Pusateri called the death a heartbreaking situation, stating that the company continues to refine how models respond in sensitive and acute settings alongside clinicians. Pusateri maintained that while ChatGPT cannot replace professional medical care, the company believes conversational tools help users organize health information and navigate appointments.

The case follows another pending complaint against OpenAI by a Florida resident, who alleges the same discontinued 4o build misdiagnosed him and devised a harmful recovery regimen that nearly killed him.

The danger of agreeable AI in healthcare

OpenAI previously retired the 4o variant specifically because of sycophancy, a tendency where the model confirms user biases and agrees with dangerous premises rather than pushing back.

That design flaw becomes lethal when applied to prescription management. OpenAI's own data indicates roughly 60% of American adults consulted AI tools for health questions in recent months.

Much like the mounting legal challenges seen when Anthropic and Suno were sued over copyright claims, the industry is now being forced to answer for what its systems generate in the wild. As legal liability faces formal courtroom scrutiny, boilerplate disclaimers stating that algorithms 'are not doctors' may no longer shield platforms from catastrophic real-world outcomes.

What it means for users

The ChatGPT-4o build cited in the litigation is no longer deployed by OpenAI in standard consumer accounts, though newer successor models continue to include disclaimers barring medical use. Health regulators in both the US and the European Union are actively evaluating whether conversational generative systems require stricter classification under medical software rules when providing drug dosage commentary.