Article · November 17, 2026 · Karolyn Law, Founder
Ontario’s Auditor General Tested 20 Approved AI Scribes. The Results Should Worry Every Clinic.
Nine fabricated information that was never said. Twelve recorded a different drug than the one prescribed. Seventeen missed mental-health details that were right there in the conversation.
AnalysisThose are not numbers from a vendor’s competitor slide, a preprint, or a conference talk. They are findings of the Office of the Auditor General of Ontario, in its Special Report Use of Artificial Intelligence in the Ontario Government, released May 12, 2026. Auditor General Shelley Spence’s team examined how the province’s 20 approved AI scribe systems performed when Supply Ontario’s procurement evaluators ran simulated doctor-patient recordings through them before the systems were made available to physicians.
Every category of error a clinician fears was in there:
- Fabrication. Nine of the 20 systems invented information, including treatment-plan steps nobody discussed, such as a therapy referral or blood tests. Evaluators also found notes stating “no masses found”, or recording anxiety, when neither was mentioned in the recording.
- Wrong medication. Notes from twelve of the 20 systems captured a different drug than the one the doctor prescribed.
- Omission. Notes from seventeen of the 20 missed key details about the patient’s mental health in at least one of the two tests, even though those details were in the simulated recordings.
The approval process did not stop them. The Auditor General found that 11 of the 20 approved vendors submitted no third-party audit reports, SOC reports or ISO 27001 certification, and five did not submit the threat risk assessments and privacy impact assessments the process required. They were approved anyway.
The province’s response, fairly stated
Ontario’s answer deserves a place in this article, because it is partly right. The errors were found during procurement testing on simulated recordings, not in live patient care. OntarioMD issued guidelines asking doctors to review the generated notes before relying on them. And the province has said it is not aware of patient harm among the more than 5,000 Ontario physicians now using these tools.
All of that can be true, and the report still matters. Here is why.
What the report actually proves
The report does not prove that AI has no place in a clinic. It proves something more specific and more useful: the safety of these systems rests on a habit, not a mechanism.
“The physician reviews the note” is the whole safety model. It depends on a tired doctor, at the end of a long day, catching a fabricated referral or a wrong drug in a fluent, confident draft that reads exactly like the thousand correct drafts before it. The Auditor General made the same point in plainer terms: doctors were not required to attest, through a sign-off in the system, that they had verified the generated notes, and the report recommends that systems enforce that attestation. That is a design gap, not a doctor gap.
The second lesson is about procurement. Twenty vendors were approved largely on their own reported data and self-attestations, some without the required independent assessments on file. When a health system buys AI on trust, it gets the accuracy it inspected for. The Auditor General made 10 recommendations to strengthen how AI is evaluated and governed. The Ministry of Health agreed with all five addressed to it; Supply Ontario agreed with four of its five and partially agreed with the fifth.
Why we are building the layer after the note
Align eCare does not write the clinical note, and this report is one reason why. Scribes write notes. Align eCare is being developed to execute the paperwork that follows the visit, and to prove it: the forms, the referral, the prescription paperwork and the billing evidence, prepared from the encounter and from sources that can be shown, not generated from a model’s confidence.
Our design responds to the Auditor General’s findings point by point:
- Rules where it counts. Billing review is designed to apply the official New Brunswick rules, and documents are designed to be assembled from the record and cited sources. Where a system might invent, ours is designed to flag and stop: a contradiction or a gap is shown to the physician, never silently resolved.
- Verification before anything moves. Under our design, nothing is transmitted on a draft. Align eCare’s proposed eAttest™ protocol binds each document to the responsible physician’s attestation of that exact version, so review is not a habit a tired clinician is trusted to remember; it is the step the document cannot move without. Verify first. Attest once. Transmit securely.
- An audit trail that exists before anyone asks. Each document is designed to carry its evidence and its attestation with it, producible on request. That is the governance the Auditor General found missing, built into the workflow rather than promised in a procurement response. Read more about our approach on the trust page, and how the platform is designed on the platform page.
None of this makes any AI system infallible, including ours, and this article would not deserve your trust if it claimed otherwise. The aim is to make errors visible, checkable, and stopped at a gate a named clinician controls, which is the standard the Auditor General’s report implicitly sets.
The work continues in New Brunswick
Align eCare is being built in New Brunswick, for the teams working to attach the patients who need care most: people relying on remote care, rural patients, unattached patients and seniors. Align eCare is in development, and a measured pilot is how we intend to put numbers on the table, in a clinic’s own results, before anyone is asked to believe a claim, ours included. If your team is evaluating AI tools this year, the Auditor General’s report is the right place to start, and the questions it raises are the right ones to ask every vendor, including us.
Sources
- Office of the Auditor General of Ontario, Use of Artificial Intelligence in the Ontario Government, Special Report, May 12, 2026. Read the report (PDF)
- CBC News, “AI transcriber for use by Ontario doctors ‘hallucinated,’ generated errors, auditor finds,” May 13, 2026.
- Related: Canada’s privacy regulators just wrote the rules for health AI
