Reports of AI chatbot medical advice contributing to patient harm surface in India

Nephrologists at Hyderabad’s National Institute of Medical Sciences reported treating patients whose conditions had worsened after they said they followed medical advice from ChatGPT rather than their prescribing doctors. In one case, a 30-year-old kidney transplant recipient reportedly stopped taking her antibiotics after a chatbot told her a normal creatinine reading meant the medication was no longer needed; her transplanted kidney’s function collapsed within weeks and she resumed dialysis. In a second case, a 62-year-old man with diabetes reportedly developed severe hyponatremia after a chatbot advised him to eliminate salt from his diet entirely.

The reports were attributed to clinicians describing what patients told them about the advice they had followed; no independent record of the specific chatbot exchanges was cited to confirm their content or wording. Doctors quoted argued that generic chatbot output misses information a treating clinician has — transplant and oncology protocols, drug interactions, a patient’s full history — and can present incorrect information with unwarranted confidence.

The cases were catalogued by the AI Incident Database alongside a separate, widely reported case of a New York patient hospitalised after substituting table salt with sodium bromide on the basis of chatbot-style advice found online. Clinicians in both cases framed the pattern as a caution about using general-purpose chatbots as a substitute for, rather than a supplement to, professional medical judgement.