Health advocates in the UK are cautioning that AI transcription tools may pose a risk to patients.

      AI scribes are increasingly being utilized in medical practices to reduce the burden of documentation, although recent discoveries indicate that these automated transcripts can cause significant errors in patient records.

      AI scribes are misrepresenting diagnoses and prescriptions

      A report from The Guardian has brought attention to how automated transcription tools are frequently mishandling medical records in the UK. According to data from Healthwatch England, an independent patient advocacy organization, patients have been identifying major transcription mistakes that threaten their health if not corrected.

      In one troubling case, an AI tool misreported an MRI summary, stating that a patient had demyelination, a serious condition associated with multiple sclerosis, when the scan actually showed "null demyelination," indicating there was no issue. Other instances documented AI tools confusing similar medication names or failing to include essential instructions for repeat prescriptions. Patient advocates warn that such inaccuracies could lead to permanent distortions in medical records, particularly if overwhelmed clinicians approve these generated transcripts without thoroughly reviewing them.

      A global issue without safety regulations

      This problem transcends the specific incidents in the UK. Recently, Australian health authorities released a warning to doctors about the risks of overly relying on unverified AI transcription software, highlighting significant privacy concerns along with clinical safety issues. In both regions, healthcare practitioners are managing numerous unregulated AI tools that can fabricate entire conversations, even including unmentioned psychiatric drugs like Prozac in patient records.

      In the United States, the situation is exacerbated as regulators have opted not to classify AI scribes as medical devices, leaving healthcare systems without centralized oversight to ensure the accuracy of these tools before deployment. Instead of saving time, these flawed AI assistants are compelling clinicians to verify each transcript for inaccuracies, undermining the intended purpose of implementing such technology.