Ontario Audit Finds Doctor AI Note-Takers Routinely Fabricate Basic Facts
The Office of the Auditor General of Ontario has released a report on 20 approved AI scribe systems used in the province’s healthcare system, and the findings raise serious concerns about AI deployment in critical domains.
Using simulated doctor-patient recordings, auditors tested the systems and had medical professionals compare AI-generated notes against the original recordings. Key findings: 12 out of 20 systems inserted incorrect drug information; 9 systems fabricated patient details or treatment suggestions that were never discussed; and 17 systems missed key mental health details discussed during the consultations.
Perhaps more troubling than the errors themselves is how these systems were selected. In the vendor evaluation process, accuracy of medical notes accounted for only 4% of the total score. By contrast, whether a vendor had a local presence in Ontario was worth 30%. Bias controls and threat/privacy assessments each accounted for just 2%. This weighting structure may have actively selected for less accurate products.
Ontario’s health ministry noted that over 5,000 physicians participate in the program with no known patient harm reports. However, the audit highlighted that no mandatory attestation feature exists — while doctors are advised to manually review AI notes, nothing forces them to do so.
This case illustrates a core challenge for AI deployment in high-stakes industries: when procurement incentives misalign with quality, the market optimizes for compliance over reliability. For the agent economy expanding into healthcare, finance, and legal sectors, this is a sobering lesson.