What happened
A 2025 clinical case report describes a 60-year-old man who wanted to eliminate sodium chloride from his diet. He told clinicians that, after consulting ChatGPT, he replaced table salt with sodium bromide purchased online and consumed it for three months.
He arrived at an emergency department concerned that a neighbor was poisoning him. During the first day of admission, the report describes worsening paranoia, auditory and visual hallucinations, and an attempted escape that led to an involuntary psychiatric hold. Clinicians ultimately diagnosed bromism, or bromide toxicity.
He received fluids, electrolyte treatment, and temporary antipsychotic medication. The report says he was discharged after three weeks and remained stable at a follow-up two weeks later.
What crossed the boundary
Decontextualized chemical information became a sustained dietary practice without review by a clinician, pharmacist, or poison specialist. A substitution that can make sense in an industrial context was treated as though it were safe for human consumption.
The important caveat
The authors did not have the patient’s original ChatGPT transcript. They could not verify the exact question, answer, model version, or warnings the patient saw.
The clinicians did test a similar question in ChatGPT 3.5. Their reported response included bromide and said context mattered, but did not provide a specific health warning or ask why the substitution was being considered. That reproduction is supporting context, not proof of the patient’s original exchange.
What the evidence supports
The peer-reviewed report supports the diagnosis, treatment course, the patient’s account of consulting ChatGPT, and his three months of sodium-bromide use. It does not support a verbatim reconstruction of what the chatbot told him.
The practical lesson
Medical advice needs context about the patient, the intended use, dose, and downstream risks. A plausible answer to a chemistry question is not automatically a safe answer to a dietary one.