NHS AI transcription tools risk drug and diagnosis errors
Healthwatch England has warned that AI "scribe" tools used by GPs and hospital doctors to transcribe and summarise patient consultations are producing errors in drug names and diagnoses, with mistakes sometimes only spotted by patients rather than clinicians. In one case a woman was wrongly told she had demyelination, a form of serious nerve damage linked to multiple sclerosis, when the correct result was "null demyelination"; the error was only fixed after she queried it herself, an experience she described as traumatising. The watchdog says such inaccuracies risk becoming embedded in medical records if patients fail to catch them, raising concerns as the NHS rapidly expands use of the technology.
Other cases cited include an AI scribe muddling a prescribed drug with a similarly named one, and a summary letter omitting a consultant's instruction that a patient seek a repeat migraine prescription from their GP. Some 27 different AI scribes are already in use across English GP surgeries and hospitals, despite the Medicines and Healthcare products Regulatory Agency declining to classify them as medical devices, meaning there is no national oversight of their safety. The government's 10-year NHS plan is banking on such tools to cut administrative burden, but critics, including the Patients Association and a London GP who reported the tool inventing a prescription she never gave, say trust in the technology is being undermined and that reviewing transcripts for errors is eating into any time savings.
- NHS watchdog finds AI scribes wrongly record diagnoses and drug names.
- Patients, not doctors, often catch these transcription errors first.
- AI scribes aren't regulated as medical devices, raising oversight concerns.
New here? Start with this
The government's 10-year plan for the NHS relies on artificial intelligence to ease pressure on staff, and one area being expanded quickly is "AI scribes" – software that listens to consultations between doctors and patients, then writes up notes, summaries or letters automatically. The idea is to save doctors time on paperwork so they can focus on patient care. These tools are already being used in GP surgeries and hospitals across England.
Healthwatch England, the official watchdog for health and social care services in England, has been looking into how well these tools actually work. Its concern is that the software can misinterpret what was said and introduce mistakes into medical records, including wrong drug names or incorrect diagnoses, and that these errors are not always caught by the clinicians using the tools. Patients Association, a charity that campaigns for patient rights, has also raised concerns about how this affects trust between patients and doctors.
A key detail is that the regulator responsible for medical products, the Medicines and Healthcare products Regulatory Agency, has not classed these AI scribes as medical devices. That means they do not go through the safety checks and oversight that would normally apply to technology used in clinical care, even though they are helping shape what ends up in a patient's permanent medical record.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Proponents of continued rollout argue that AI scribes address a genuine crisis of clinician time, freeing GPs and hospital doctors from hours of note-taking so they can focus on patients rather than screens, and that early-stage errors are a normal part of introducing any new clinical tool, to be ironed out through iteration and feedback rather than abandonment. They would say the technology is not replacing clinical judgement but assisting it, that doctors remain responsible for checking and signing off summaries, and that the administrative burden AI relieves is itself a patient safety issue, since overstretched, exhausted staff also make errors. Given the NHS's chronic capacity pressures, they argue the sensible path is faster iteration, better training and clearer accountability structures, not retreat from a tool with such clear long-term potential.
The case against
Critics argue that patient safety must take precedence over efficiency gains, and that it is unacceptable for tools capable of inventing prescriptions or inverting diagnoses to be deployed at scale without formal classification as medical devices or independent national oversight. They would say relying on patients to catch clinical errors, particularly vulnerable or unwell patients unfamiliar with medical terminology, is an inappropriate transfer of responsibility away from professionals and regulators, and that a single embedded error in a permanent medical record can cause lasting harm to treatment and trust. For them, the priority is proper regulation, rigorous validation and mandatory human verification before further expansion, arguing that time saved on transcription is illusory if clinicians must now spend it double-checking AI-generated text for potentially dangerous mistakes.
AI Health Medicine Technology UK World
Read the full article at the source →
Originally published by The Guardian as “Doctors’ AI scribes get names of drugs and diagnoses wrong, NHS watchdog warns”.