A 60-year-old man has been hospitalised with bromide intoxication a rare condition largely forgotten for decades after following medical advice sourced from ChatGPT, according to a report published in the Annals of Internal Medicine.

The patient’s ordeal began when he developed suspicions that his neighbour was secretly poisoning him. Concerned about the health effects of sodium chloride (common table salt) after reading online articles, he consulted ChatGPT for alternatives. Acting on the chatbot’s suggestions, he replaced salt with sodium bromide a decision that ultimately triggered bromism, a condition known to cause serious neuropsychiatric symptoms, including psychosis and hallucinations.

“He was noted to be very thirsty but paranoid about water he was offered,” the report noted. The man resorted to distilling his own water and imposing severe dietary restrictions. Once admitted to hospital, his paranoia and hallucinations intensified, leading to an involuntary psychiatric hold after he attempted to escape.

Bromism was once a recognised medical problem in the late 19th and early 20th centuries, when bromide salts were prescribed for neurological disorders and insomnia. Chronic use was later found to cause nervous system damage, and the U.S. banned bromides from over-the-counter medicines in 1975.

The medical team could not review the man’s original ChatGPT conversation, but test queries revealed the AI still suggested bromide as a substitute for chloride without offering a specific health warning or asking for context, as a human doctor might.

While ChatGPT and other AI tools have occasionally aided in detecting medical issues including one widely publicised case earlier this year where a mother used the chatbot to identify her son’s rare disorder experts warn that such outcomes are exceptions.

“The ability of ChatGPT (GPT-4.5 and GPT-4) to detect the correct diagnosis was very weak for rare disorders,” noted a study in the Genes journal, emphasising that AI cannot replace thorough in-person evaluation.

Medical professionals stress that AI-driven advice should only be considered when integrated into clinical settings and supervised by certified healthcare providers.