ChatGPT Advice Leads to Rare Bromide Poisoning in 60-Year-Old Man
A 60-year-old man’s quest to “improve his health” by reducing dietary chloride took a dangerous turn after he sought guidance from ChatGPT. Three months later, he was in a hospital emergency department with paranoia, hallucinations, and other troubling psychiatric symptoms. Doctors ultimately diagnosed him with bromism, a rare form of poisoning caused by prolonged exposure to bromide compounds.
The man’s case, published Aug. 5 in Annals of Internal Medicine Clinical Cases, began with his research into the health effects of table salt, also known as sodium chloride.
“After reading about the negative effects that sodium chloride, or table salt, has on one’s health, he was surprised that he could only find literature related to reducing sodium from one’s diet. Inspired by his history of studying nutrition in college, he decided to conduct a personal experiment to eliminate chloride from his diet. “
According to the report, the AI allegedly suggested replacing sodium chloride with sodium bromide — a chemical once used in sedatives, anticonvulsants, and sleep aids but largely abandoned decades ago due to toxicity concerns. While sodium bromide may appear chemically similar to table salt, it can be harmful when ingested over time. The man purchased the substance online and began using it in place of salt in his meals.
Bromide-containing drugs were common in the late 19th and early 20th centuries, but by the 1970s and 1980s, U.S. regulators removed most from over-the-counter shelves. The decision followed recognition that chronic exposure could cause bromism, a toxic buildup of bromide in the body. Symptoms can include agitation, psychosis, mania, delusions, memory problems, coordination issues, and skin reactions.
Bromide disrupts neuron function when levels in the body climb too high. In severe cases, it can mimic or mask other conditions — as happened with this patient. His initial lab tests suggested elevated chloride, but further analysis revealed a false reading known as pseudohyperchloremia, caused by bromide interfering with standard blood chemistry measurements.
When he arrived at the emergency department, the man reported that his neighbor was trying to poison him. Tests showed high blood alkalinity and carbon dioxide buildup. Over his hospital stay, paranoia intensified and hallucinations began. He attempted to leave the facility and was placed under an involuntary psychiatric hold, during which he was started on antipsychotic medication.
Once stabilized with fluids and electrolyte monitoring, he revealed his use of sodium bromide as a salt substitute, mentioning other symptoms that had emerged in recent weeks — insomnia, fatigue, clumsiness, severe thirst, acne, and small reddish skin bumps. These clues reinforced the bromism diagnosis.
Over three weeks, doctors tapered his antipsychotic medication, and he was discharged once stable. At a two-week follow-up, his condition remained improved.
While the exact conversation between the man and ChatGPT wasn’t available to the researchers, he said the AI suggested bromide as a substitute for chloride.
The authors tested a similar query on ChatGPT-3.5 and found that it did include bromide in its response, adding a vague “context matters” note but failing to warn against consumption or seek further clarification.
“While it is a tool with much potential to provide a bridge between scientists and the nonacademic population, AI also carries the risk for promulgating decontextualized information,” the report authors 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. Thus, as the use of AI tools increases, providers will need to consider this when screening for where their patients are consuming health information.”
When contacted for comment, OpenAI directed reporters to its terms of service, which explicitly state that outputs are not a substitute for professional medical advice and should not be relied upon as the sole source of factual information. A company spokesperson noted that OpenAI’s safety teams work to reduce such risks and encourage users to consult qualified professionals.
In a broader warning, a separate team of researchers recently tested six AI models, including ChatGPT, on their ability to interpret doctors’ clinical notes. They found that these systems were vulnerable to “adversarial hallucination attacks,” producing convincing but false clinical details. Technical safeguards reduced errors but didn’t eliminate them.
For the patient in this case, a seemingly simple experiment to cut chloride from his diet, aided by AI, nearly ended in disaster. While bromism is rare today, the ease of buying industrial chemicals online and the growing role of AI in everyday decision-making may mean such cases won’t remain isolated.
Moumita Basuroychowdhury is a Contributing Reporter at The National Digest. After earning an economics degree at Cornell University, she moved to NYC to pursue her MFA in creative writing. She enjoys reporting on science, business and culture news. You can reach her at moumita.b@thenationaldigest.com.
