In NHS records the scribe's error is usually caught only by the patient

The error that enters a patient record

In the cases Healthwatch England collected, a scribe left out the word that reversed the meaning of a test result reading "null demyelination", and the patient was told they had demyelination. In another consultation the drug the doctor had prescribed was swapped for one with a similar name. A summary letter left out the instruction to ask a GP for a repeat migraine prescription; in another patient record a Prozac prescription appeared that was never discussed. All four were written into the durable patient record.^[1]

27 separate instances of these tools run across the health service in England and none is regulated as a medical device. The guidance the MHRA published in August draws the line plainly: a system that only transcribes what was said falls outside the device regime, while a system that suggests a diagnosis or a treatment can fall inside it. The test looks at the function a vendor describes rather than at the tool's name. Yet a tool that writes the wrong drug name into a record alters the text carrying a doctor's decision even with no diagnostic function at all. Healthwatch England reports that patients are usually the ones who notice these errors, which leaves the checking of a wrong record to the patient's own attention.^[1]

Where the review never runs

In Ireland the Digital Contact Centre inside the Department of Justice's immigration unit runs a chatbot built on Microsoft Copilot alongside ID Pal, an identity check that uses machine learning. What the ICCL established through Freedom of Information requests is that no tender process and no risk assessment were carried out for that set-up. The shared constraint across the two tools stands here: both produce official records about people, and in both the review that tests a tool before deployment never runs. In England that outcome comes from the device test; in Ireland it comes from an assessment nobody carried out.^{[1], [2]}

The ICCL said the shortcomings "present a real risk of harm to some of the most vulnerable people in our society". The department has published no detailed account of how far the tools feed into application outcomes. Without that account it cannot be established from outside whether the chatbot only supplies information or takes part in preparing a decision; what the Freedom of Information requests surfaced is the absence of the process itself.^[2]

What would close the gap?

Healthwatch England's recommendation is modest and testable for exactly that reason: tell the patient a scribe was used, and let the patient see the note before the patient record is finalised. Accuracy stays where it was; what changes is that the party already catching the errors gains a chance to act before the patient record freezes. That is also the measurable part: a national document naming which tools are in use and stating the point at which a patient sees the note opens a route to oversight without waiting for the classification argument to be settled.^[1]