A Seattle man is currently recovering after a bizarre and dangerous health incident involving AI advice. Looking for a way to cut table salt from his diet, the 60-year-old turned to ChatGPT. The chatbot suggested replacing sodium chloride with sodium bromide, a chemical substance that was common in early 20th-century medicine but is now known for severe toxicity.
Over the course of three months, the man followed the advice and began experiencing intense paranoia and hallucinations. He was eventually hospitalized after becoming convinced his neighbors were poisoning him. Upon arrival, medical staff initially struggled to identify the source of his illness. Blood tests revealed an unusual electrical imbalance in his body that misled the clinical team.
After consulting with poison control, doctors realized the man had developed bromism, or bromide poisoning. His blood levels of bromide were 230 times higher than the normal range. The substance had entered his brain, disrupting the electrical activity of his neurons and triggering his psychotic state. He spent three weeks in the hospital receiving treatment to manage his electrolytes and address his neurological symptoms.
This case highlights a growing concern in the medical community regarding the use of AI for specialized health advice. When doctors tested the same prompt, the chatbot once again recommended sodium bromide without including any warnings about its toxicity or the extreme health risks involved. The situation serves as a stark reminder that generative AI lacks the capacity for medical judgment and that independent professional verification is vital for any health-related decision.