Medical researchers are urging the public to use caution when seeking health advice from artificial intelligence (AI) chatbots, after a man developed a rare neurotoxic condition following a conversation with ChatGPT about removing table salt from his diet.
As researchers noted, bromism was “well-recognized” in the early 20th century and thought to have accounted for almost one in 10 psychiatric admissions at that time, largely due to the widespread use of bromide in over-the-counter sedative drugs. Its occurrence plummeted after the Food and Drug Administration restricted bromide salts in ingestible products.
According to the paper, the patient told doctors he decided to eliminate sodium chloride—the essence of table salt—from his diet after reading about chloride’s potential negative health effects. Following a consultation with ChatGPT, he spent the next three months replacing sodium chloride with sodium bromide he bought online—despite his having read that “chloride can be swapped with bromide, though likely for other purposes, such as cleaning.”
The man later arrived at a hospital claiming that his neighbor was trying to poison him and describing multiple dietary restrictions. He distilled his own water at home and, though thirsty, was noted as being paranoid about drinking any water offered to him.
With no prior psychiatric history, the patient became increasingly paranoid while in the hospital and reported both auditory and visual hallucinations. He tried to escape the hospital and was involuntarily admitted to the psychiatric unit, where he was treated with fluids, electrolytes, and antipsychotics.
Once stabilized, he reported other symptoms consistent with bromism, including facial acne, poor muscle control, excessive thirst, and insomnia.
According to the paper, his antipsychotic medication was gradually tapered over three weeks before he was discharged. A follow-up two weeks later found he remained stable.
The authors of the paper said the case shows “how the use of artificial intelligence can potentially contribute to the development of preventable adverse health outcomes.” They particularly warned about the risks of AI chatbots providing information that lacks proper context.
The AI company stressed, however, that ChatGPT is not a substitute for professional help and that users should instead think of it as a “partner to help you understand results, ask the right questions in the time you have with providers, and weigh options as you make decisions.”
The authors of the paper, which was published before the release of GPT-5, said they were unable to access the patient’s ChatGPT conversation log and therefore could not determine exactly what advice he had received. They noted the patient had apparently been using an older version of ChatGPT, likely GPT 3.5 or 4.0.
When the researchers consulted ChatGPT 3.5 about what could replace chloride, the AI mentioned bromide without providing a health warning or asking why the authors were seeking such information, “as we presume a medical professional would do,” the paper stated.
“It is highly unlikely that a medical expert would have mentioned sodium bromide when faced with a patient looking for a viable substitute for sodium chloride,” the researchers wrote.
