Healthwatch Warns AI Scribes Are Misrecording Drugs and Diagnoses

Healthwatch England has warned that artificial-intelligence tools used to transcribe medical consultations are introducing errors into patient records, including incorrect drug names and diagnoses.

The statutory patient watchdog said patients had identified multiple mistakes that health professionals did not catch. In one case, an AI-generated summary incorrectly recorded that a patient had demyelination, a form of nerve damage, when the intended wording was “null demyelination.” The hospital corrected the record after the patient raised the error.

Healthwatch also cited a case in which an AI scribe confused a prescribed medicine with another drug with a similar name. In a separate incident, an automatically generated letter omitted a consultant’s instruction that a patient should request a repeat migraine prescription from a general practitioner.

The findings highlight a central risk as the National Health Service in England expands the use of ambient AI systems. These tools listen during consultations and generate notes or summaries, with the aim of reducing administrative work and allowing clinicians to spend more time with patients.

Healthwatch said mistakes may remain in medical records when patients do not notice them. It called for clearer procedures explaining how people can report and correct inaccuracies produced by AI tools or by the professionals using them.

The watchdog also raised concern that Britain’s Medicines and Healthcare products Regulatory Agency does not classify AI scribes as medical devices. Healthwatch said that decision leaves no nationwide oversight system in England to assess whether the products are safe and effective.

According to the report, doctors in general practice and hospitals in England are already using 27 different AI-scribe products. Previous warnings have included the possibility that the NHS and clinicians could face legal action over errors, as well as concerns that transcription systems may struggle with accents or insert information that was never discussed.

Researchers and clinicians cited in the report stressed that human-produced notes are not error-free either. A survey of 1,003 UK general practitioners found that more than half believed AI-generated records were more accurate than their own. However, clinicians have also said that reviewing every automated transcript for mistakes can reduce the time savings promised by the technology.

The warning comes as the British government’s 10-year health plan places AI scribes within a wider effort to shift the NHS from analogue administration toward digital services. Healthwatch’s evidence suggests that rollout will need to be accompanied by reliable clinical review, transparent accountability and a simple way for patients to challenge errors before they affect care.