A man’s attempt to improve his diet using AI-generated health tips ended with him in the hospital, suffering from a rare and almost forgotten form of poisoning.

How It Started: A Search for a Salt Alternative

The patient, whose identity has not been disclosed, set out to lower his sodium chloride (table salt) intake. Like many people, he turned to an AI chatbot—specifically ChatGPT, for suggestions.

Instead of recommending a common low-sodium alternative like potassium chloride, the chatbot suggested sodium bromide, a compound better suited for sanitizing hot tubs than seasoning food.

The man ordered the chemical online and began incorporating it into his meals.

From Diet Change to Delusions

Three months later, he arrived at an emergency department exhibiting severe psychiatric symptoms, including paranoid delusions that his neighbor was trying to poison him.

During the first 24 hours of his hospital stay, his paranoia escalated into both auditory and visual hallucinations. When he tried to escape, doctors placed him on an involuntary psychiatric hold for grave disability.

Once stabilized with antipsychotic medication, the man revealed his AI-inspired dietary change. Blood tests confirmed the cause: bromism, a toxic buildup of bromide in the body.

Understanding Bromism

• Normal bromide levels in healthy people: under ~10 mg/L.
• This patient’s level: 1,700 mg/L—170 times higher than normal.
• Symptoms: Confusion, hallucinations, paranoia, poor coordination, and, in severe cases, coma.

Bromism was once relatively common, especially in the early 20th century, when bromide salts were widely used in over-the-counter sedatives and medications. At one point, it was estimated to account for up to 8% of psychiatric hospital admissions.

The condition virtually disappeared in the 1970s and 1980s, when bromide-containing drugs were phased out.

Treatment and Recovery

Doctors treated the man over the course of three weeks, flushing the bromide from his system. He was discharged with no major long-term health effects.

The Bigger Issue: AI and Medical Advice

While bromism itself is rare today, the case underscores a modern risk: relying on AI tools for medical guidance without expert oversight.

The authors of the case study, published in the Annals of Internal Medicine: Clinical Cases, warned that AI models can:

• Generate scientific inaccuracies.
• Lack the ability to weigh medical context or risks.
• Spread misinformation in a convincing tone.

“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 authors wrote.

What This Means for AI Users

Generative AI can be useful for everyday queries—finding a recipe, drafting an email—but it should not be treated as a substitute for licensed medical professionals.

Key takeaway: If a health recommendation sounds unusual, double-check it with a trusted doctor or pharmacist before acting on it.