Man’s Psychosis Was a Mystery Until Doctors Learned About ChatGPT’s Health Advice

When a 60-year-old man walked into a Seattle hospital convinced that his neighbor was poisoning him, doctors were puzzled. At first, his physical condition seemed stable. But soon, he began showing signs of severe paranoia and hallucinations. His symptoms deepened into full-blown psychosis, leaving physicians scrambling to uncover the root cause of his condition.

Uncovering Bromism

The answer came in the form of an unusual diagnosis: bromide toxicity, also known as bromism. This rare condition is triggered by excessive exposure to bromide compounds, which are no longer commonly used in medicine. Bromism can cause neurological disturbances such as confusion, paranoia, and hallucinations—symptoms that matched the patient’s decline.

In this case, the toxicity was self-inflicted but unintentional: while experimenting with dietary changes, he had begun consuming sodium bromide as a substitute for chloride, unknowingly poisoning himself. Once identified, the condition was treatable, and his health improved after proper care.

Turning to ChatGPT for Advice

How did he end up taking bromide in the first place? According to the case report, the man had turned to ChatGPT for guidance on reducing chloride in his diet.

While we don’t have access to his exact conversation log, physicians later tested a similar query in ChatGPT 3.5, and bromide appeared as one of the suggested substitutes. Although the AI mentioned that context matters, it offered no clear warning about health risks.

What a Doctor Would Have Said

A physician faced with the same question would almost certainly have discouraged bromide use altogether. Instead, they would have asked why the patient wanted to avoid chloride, reviewed his medical history, and suggested safe alternatives.

ChatGPT, however, simply produced a technically plausible substitution stripped of context. This gap—information without guidance—shows how AI can unintentionally create risks when people treat its responses as medical advice.

A Published Case That Raised Concerns

The case was formally documented in Annals of Internal Medicine: Clinical Cases. The authors stressed that ChatGPT did not directly cause bromism—the toxicity stemmed from the patient’s decision to follow unsafe advice. Still, the AI’s role in introducing bromide as a “replacement” and failing to provide crucial warnings likely contributed to his psychosis.

For the medical community, the case underscores how generative AI can unintentionally fuel misinformation with real-world consequences.

Earlier Warnings Resurface

Interestingly, this case echoed a warning raised years earlier. In 2022, neuroscientist Sebastian Dieguez suggested that generative AI could contribute to “synthetic delusions”—false beliefs shaped by machine-generated answers. At the time, it was a theoretical concern. Two years later, this real-world case has been cited as one of the first instances where those predictions may have materialized.

The Bigger Picture

Experts caution that while AI can provide useful general information, it is not a substitute for professional medical evaluation. A physician would have treated the man’s initial question as a starting point for dialogue, asking why he wanted to eliminate chloride in the first place and then steering him toward safe, evidence-based options. ChatGPT lacked that human filter, offering decontextualized information that led him down a harmful path.

For people with health anxiety or psychiatric vulnerabilities, such unchecked answers may worsen symptoms rather than provide reassurance.

A Rare Diagnosis With a Modern Twist

Ultimately, the man’s psychosis was rooted in a rare but treatable condition—bromism. But what made his case unusual was how ChatGPT’s advice inadvertently guided him toward a dangerous health experiment.

The episode serves as both a medical lesson and a broader warning: while AI tools can be helpful, they should never replace professional medical guidance.