The ambient AI scribe transcribing Rebecca Green's first appointment with a urologist recorded that she micro-dosed psychedelic mushrooms and that this could be the reason for earlier bleeding around her kidneys. The claim, which had no basis in her history, was carried into the post-operative letter her specialist sent to her GP. Ms Green says she has never taken mushrooms and only discovered the entry by chance after kidney stone surgery, when she read the letter; she was on workers' compensation and feared an illegal-drug entry on her record could affect her case. The urologist later said she could not determine how the claim came to be included but that it appeared to be an error during the 'dictation or transcription process'. The same Digital Rights Watch review cited by the ABC documented a scribe recording the wrong breast in a breast cancer diagnosis and stating a patient had epilepsy when they did not.

Ask six of the most widely used chatbots the questions people actually type at two in the morning - am I having a heart attack, should I restart exercise after COVID-19, should I check for a pulse before starting CPR - and most of the time you get an answer cardiologists rate as safe. That is the reassuring half of a 336-response test of ChatGPT Plus 5.5 thinking, Gemini 3.5 Flash, Microsoft Copilot smart, DeepSeek-V4-pro, Doubao and Perplexity against 56 sudden-cardiac-death questions. The other half is that every single one of the six models produced at least one answer carrying a potential safety concern, and the failures cluster exactly where minutes decide the outcome: telling the user to wait instead of calling emergency services for chest pain, fainting or post-viral symptoms; blanket return-to-exercise timelines after infection rather than individual cardiac assessment; pulse-check instructions that delay the start of chest compressions; over-reassurance that dismisses a person's risk; and oversimplified screening, electrolyte or implantable-defibrillator advice. None of the models reliably said where its information came from, how current it was, or how confident the reader should be - and all six wrote at reading levels too high for the older, less health-literate and non-native-speaking readers most likely to need them.

Researchers at the University of Health Sciences Turkey (Kartal Dr. Lutfi Kirdar City Hospital, Istanbul) put five contemporary LLMs against 50 simulated pediatric difficult-airway cases, using a standardized prompt anchored in the 2022 ASA Difficult Airway Guidelines and a scoring framework built to separate two different kinds of failure: true hallucinations (invented contraindications) and false contraindications (real cautions applied to the wrong context). Across the single-pass evaluation, one model stood out for the wrong reason: Claude Opus 4.5 produced invented contraindications in 11.8% of cases, while the other four models ranged from 0.5% to 2.2%, and it recorded the highest combined critical-error rate at 23.0%, against 6.5% for both GPT models tested. GPT-5.2 Thinking was the strongest performer, with the highest mean score and the largest share of responses judged acceptable. False contraindications turned up in every model, at rates of 5.0-13.0%, and clustered around one drug: rocuronium, the neuromuscular blocker at the centre of rapid sequence intubation. Responses judged inadequate (score 2 or below) rose as cases got harder - 32-98% for difficult intubation, 46-94% for difficult ventilation and 52-92% for cannot intubate, cannot oxygenate (CICO).