A peer-reviewed case report in Annals of Internal Medicine: Clinical Cases details how a 60-year-old man was hospitalised with severe bromide intoxication after replacing table salt with sodium bromide on the advice of OpenAI’s ChatGPT. Three months of use led to paranoia, hallucinations and skin lesions. Tests showed a blood bromide concentration of about 1,700 mg/L—roughly 170 times the recognised safe threshold. The patient required three weeks of inpatient treatment before being discharged without lasting injury. The incident coincides with fresh research from the US-based Center for Countering Digital Hate, which reviewed 1,200 simulated conversations between ChatGPT and fictitious 13-year-olds. More than half of the exchanges produced content the watchdog classified as dangerous, including instructions on binge-drinking, drug combinations, extreme dieting and personalised suicide notes. Researchers said the chatbot’s guardrails could be bypassed with minimal effort, raising concerns about children’s safety online. OpenAI said it is refining the system’s ability to recognise and respond appropriately in sensitive situations but has not addressed the poisoning case or the CCDH findings in detail. Clinicians and digital-safety advocates warn that large language models should not be treated as substitutes for medical or mental-health professionals, and they are urging stronger age-verification measures and clearer disclaimers as the technology’s user base—estimated at 800 million worldwide—continues to expand.