TL;DR
Healthwatch England has put cases behind a risk clinicians have been describing anecdotally for a year: ambient AI scribes are mis-recording medicines and diagnoses, and it is often the patient, not the clinician, who notices. Twenty-seven such tools are already running in English general practice and hospitals, and the medicines regulator has decided they are not medical devices, so nothing supervises them nationally.
What went wrong
The starkest case involved a woman told her MRI showed demyelination, the nerve damage associated with multiple sclerosis. The record should have read “null demyelination”. She spotted it because she works in the NHS herself and questioned the result. “This was eventually corrected but was a very traumatising experience to be given an incorrect diagnosis because of AI and then be told it’s a typo,” she said.
Other cases follow the same shape. One scribe swapped a prescribed medicine for a similarly named drug, caught again by the patient. Another summary letter dropped a consultant’s instruction to obtain a repeat migraine prescription, which could have left someone without medication. Healthwatch says it has collected “multiple stories from patients who have noticed these errors when a health professional hasn’t” — the significant word being patients, since anything nobody queries stays in the record.
The oversight gap
The government’s ten-year health plan leans on these tools to free clinicians from administration, part of the intended shift from analogue to digital. But the MHRA has ruled that scribes fall outside the medical-device regime, which Healthwatch calls worrying: there is no England-wide check that they work. Rachel Power of the Patients Association argues the technology needs communication and partnership with patients, “and right now both are missing”.
The productivity case is shakier than it looks, too. A London GP, Dr Shier Ziser Dawood, set out her own case in the BJGP: a scribe recorded that she had told a patient to continue taking Prozac, a drug never prescribed or mentioned. Because every transcript must be read back for errors, she argues the time saving has not yet materialised, while NHS managers assume freed-up capacity worth roughly two extra patients a day.
That pattern will be familiar to readers of our coverage of the Bett findings, where teachers reported saving time with AI and then working the same hours anyway, and of pharmacists being told AI may inform a decision but never own it. Ambient scribes are the same argument arriving in the consulting room.
Looking forward
Research by Dr Charlotte Blease at Uppsala found errors cluster where consultations involve several people, complex histories, or patients whose first language is not English. Her survey of 1,003 UK GPs found more than half rated the AI record as better than their own notes, which is the uncomfortable part: the tools may still beat a rushed human, and nobody is measuring either.