An AI transcription tool erased one word from a test result and gave a patient a diagnosis they never had.
Healthwatch England, the statutory patient watchdog, has warned that AI scribes used to transcribe clinical notes are introducing errors into NHS records, including wrong drug names and wrong diagnoses. In one case cited in the warning, a patient was told they had demyelination, the nerve damage linked to conditions like multiple sclerosis. The actual test result read "null demyelination" - no such damage at all. The AI tool had simply dropped the word "null," flipping the result from reassuring to alarming.
This matters because these tools are being adopted as time-saving scribes, not as diagnostic software, so nobody is reviewing their output with the scrutiny a diagnosis deserves. A single missing word in a note can cascade into unnecessary tests, wrong prescriptions, or a patient carrying a diagnosis they don't have. Transcription errors are old news in medicine; what's new is doing them at AI speed and AI scale, across a health system already short on staff to double-check the machine's homework.
Spellcheck never had this problem, because spellcheck never claimed to understand what it was reading.