Ontario Audit Flags 20 AI Scribe Vendors for Incorrect, Hallucinated Patient Notes
Updated
Updated · Ars Technica · May 14
Ontario Audit Flags 20 AI Scribe Vendors for Incorrect, Hallucinated Patient Notes
9 articles · Updated · Ars Technica · May 14
Twenty AI medical scribe vendors pre-qualified by Ontario failed simulated transcription tests, with the auditor general finding every product produced inaccurate or incomplete patient notes in at least one case.
Nine vendors hallucinated patient information, 12 recorded details incorrectly, and 17 missed key mental-health information from two mock doctor-patient conversations reviewed in the audit.
The report cited errors with direct care implications, including invented referrals for blood tests or therapy, wrong prescription names, and omitted discussion points that could distort treatment plans.
Ontario had recommended the tools to healthcare providers as overworked doctors increasingly use AI scribes to turn consultations into structured records, widening concern about patient-safety risks from government-approved systems.
Can saving doctors from burnout justify the risk of AI scribes inventing patient facts and medical records?
If government-approved AI harms a patient, who is to blame: the doctor, the vendor, or the government?
Critical Audit: Ontario’s AI Scribes Plagued by Inaccuracies, Privacy Risks, and Low Staff Training
Overview
Ontario's Auditor General recently uncovered critical flaws in the province's AI scribe systems, which have been used in healthcare since 2023. The audit found that, although doctors are told to manually review AI-generated notes, there are no strong IT controls to enforce this or require formal confirmation. This lack of enforced verification puts patient care at risk. Additionally, concerns about security and privacy remain, as Supply Ontario disagreed with increasing the importance of these factors in procurement. These systemic weaknesses highlight the urgent need for better oversight and stronger safeguards to protect patient safety and data integrity.