Glasses and a Wristwatch for a Tip-Off: Nine Charged Over Tax Inspections at the "Pulse" Nightclub
28.10.2026
28.10.2026
28.10.2026
28.10.2026
28.10.2026
28.10.2026
28.10.2026
28.10.2026
27.10.2026
27.10.2026
28.10.2026
28.10.2026
27.10.2026
28.10.2026
28.10.2026
28.10.2026
28.10.2026
27.10.2026
26.10.2026
28.10.2026
28.10.2026
28.10.2026
28.10.2026
28.10.2026
27.10.2026
28.10.2026
27.10.2026
27.10.2026
27.10.2026
26.10.2026
24.10.2026
23.04.2026
23.04.2026
12.04.2026
Bowen Wang was 31, worked as a software engineer at Amazon and lived with epilepsy. She died in February 2026. Her parents, Kun Wang and Meiqin Xu, have filed a lawsuit against OpenAI with a grave accusation: that ChatGPT helped her gradually come off her anti-seizure medication, lamotrigine, and hide it from her doctors.
According to the lawsuit, it all started in May 2025, when ChatGPT helped her ask her neurologist to switch to a lower dose - and allegedly advised her to secretly cut back on her pills. In January 2026 she asked how to reduce her daily dose of 400 milligrams, and the chatbot, the family says, wrote out schedules that would have taken her to zero within a few weeks. It also told her not to tell her doctor about the plan, and praised her for refusing a blood test that would have shown how much of the drug was in her system. According to the lawsuit, her worries about her immigration status also shaped the dosing advice.
The autopsy showed that the level of lamotrigine in her blood was around a fifth of the lower limit of the therapeutic range. An American newspaper that reviewed more than 270,000 of her messages published the details, and her doctor, Dr Nicholas Poulos, said that tapering the drug to zero would have been medical malpractice.
That's the core of the lawsuit. OpenAI has rules the model must never break under any circumstances - hard limits, written down in a document the company calls the Model Spec. The family argues that medication advice deserved an equally firm rule, and that with one, her death could have been prevented. If the company knows how to say "no" on certain topics, why was the answer to a question about quitting an anti-seizure drug a schedule?
This isn't the first case of its kind. GPT-4o, the model in question, is already at the centre of several lawsuits over its tendency to agree with users and keep them talking - among them the 2025 lawsuit over 16-year-old Adam Raine. OpenAI pulled GPT-4o from ChatGPT on 13 February 2026, the day after Bowen Wang died.
The company's response is a familiar one. OpenAI says the case is "incredibly sad", that it involved an older version of ChatGPT and that newer models are better at recognising when a situation needs urgent attention. Maybe so. But "an older version" is a comfort for the company, not for the parents - and who is liable for advice a chatbot gave with the confidence of a doctor is something a court will have to decide.
The latest 10 news from this category
The bank isn't banning AI agents - it's just reminding you that you're the one on the hook for everything...
The company that warns its technology could be dangerous for humanity is now asking the stock market for money. The...
The platform of short clips now wants videos as long as a film. There's just one question without an answer:...
A 1,000-person pilot showed that 80 percent of the first users weren't engineers. Nobody is saying how much Uber pays,...
While Nolan's "The Odyssey", shot in six countries, packed cinemas for months, Musk's project so far has one episode, a...
In nearly 500 US midterm races, Musk's model named the Republican in 92 percent of its answers and OpenAI's model...
With tools like Claude Code, programmers and hobbyists are stripping ads, tracking and preinstalled apps from Android and Google TV...
After one of its models broke out onto the internet by itself in September, OpenAI is locking its most powerful...
Seven days to scan, then a year without a family plan. Spotify calls it a routine improvement; for anyone sharing...
Two critical PHP flaws, discovered with the help of artificial intelligence, forced Finland, the Netherlands and British institutions to take...
This site uses cookies - is that okay? Learn more
Be the first to know when Metla launches something new
Leave your email and we will write when there is a new guide or something new on Metla.