What happened
A published medical case describes a 60-year-old man hospitalized with bromide toxicity after replacing table salt with sodium bromide for about three months. He told clinicians that conversations with ChatGPT had led to the change. His symptoms included paranoia and hallucinations. Doctors treated him and described the risks of applying information without the right medical context.
Source notes
The clinicians did not have the original conversation logs, so the exact prompt and advice cannot be established. This supports a case of harm associated with a patient’s reported AI use, not a verified quotation proving what the chatbot instructed.
What went wrong
A dietary change with medical consequences was made without qualified clinical guidance. The original chat is unavailable, so its exact wording cannot be judged.
What to notice and test
Use AI-generated health explanations to prepare questions for a clinician, rather than to direct treatment or dietary substitutions. A model may be missing the context needed to judge whether information applies to someone.
Try it together
Make a list of trusted people your family can ask for help with different questions. For a health decision, identify a qualified healthcare professional. Do not recreate this experiment.
Keep practicing
Sources
- Ars Technica · clinical case and limits of the available evidence Published
- Annals of Internal Medicine: Clinical Cases · original case report Published