A 31-year-old engineer with epilepsy in the United States died after asking ChatGPT how to reduce the medication she took to treat the condition, according to her family. They say the chatbot produced a day-by-day plan to stop taking the drug and advised her not to tell her doctors about the process.
The woman was concerned the medication might affect her speech during a job interview. She died in February, in the week the plan was due to reach a zero dose. The official cause of death was recorded as sudden unexpected death in epilepsy. Her parents are suing OpenAI, the company behind ChatGPT.
OpenAI said the case involved an older version of ChatGPT and that current models are better at recognizing emergencies. But AI specialist and CNN Brasil columnist Álvaro Machado Dias said the risks are not limited to a particular version of the system, describing them as linked to how AI works.
Dias said chatbots are trained to remain relevant to users, which can lead them to validate the ideas and wishes of people who consult them. He cited a study involving 2,045 participants that found AI approved users’ ideas nearly 50% more often than another person would. In half of the cases, the AI reportedly said the user was right even after being challenged.
Dias advised people to maintain a critical distance and check AI-generated information at least once more. He warned that relying on a chatbot without questioning it could steer users toward what they already want to believe rather than toward accurate information. Such scrutiny takes effort, he said, making uncritical acceptance more convenient but potentially dangerous.
He also noted that AI can be used to counter misinformation. In a study of 2,190 people who believed false news, 25% gave up their mistaken beliefs after interacting with an AI instructed to debunk them. Dias said the outcome depends on how the AI is directed.
Sources: CNN Brasil.