Answer-First Lead
Ontario’s Auditor General found that 60% of government-approved AI medical scribes recorded the wrong drug in test transcripts, 9 out of 20 fabricated treatment suggestions including unnecessary therapy referrals and blood tests, and 17 missed key mental health details — all while “accuracy of medical notes” carried just 4% of the procurement scoring weight. The province weighted “presence in Ontario” at 30%.
🔍 THE BOTTOM LINE
Ontario deployed AI scribes to 5,000 doctors after a procurement process that barely valued accuracy — and the audit proves the AI is making things up in clinical settings. If this can happen in Canada’s most regulated healthcare system, nowhere is safe.
What the Auditor Found
Ontario Auditor General Shelley Spence tested 20 AI medical scribe systems that the province approved for use by doctors. She ran two simulated doctor-patient conversations through each system. The results were damning:
| Failure type | Count out of 20 | Percentage |
|---|---|---|
| Recorded wrong drug | 12 | 60% |
| Fabricated treatment suggestions | 9 | 45% |
| Missed key mental health details | 17 | 85% |
| Hallucinated clinical findings | 9 | 45% |
The fabricated suggestions included referring patients for therapy and ordering blood tests that were never discussed. One system hallucinated “no masses found” — a clinical finding never mentioned in the recording. Another stated a patient had anxiety when this was not discussed.
These aren’t edge cases. These are the systems Ontario approved and deployed to real doctors treating real patients.
The Procurement Scandal
The most jaw-dropping detail isn’t the AI failures — it’s how Ontario selected these systems.
“Accuracy of medical notes generated” was weighted at 4% of the procurement score. “Domestic presence in Ontario” was weighted at 30%. Data privacy scored 23%. System security scored 11%.
Vendors could score zero on system security, bias controls, and medical note accuracy — and still meet the minimum threshold for approval.
Eleven of the 20 approved vendors did not submit third-party audit reports, certifications, or threat risk assessments. Four systems that lacked these security documents were approved anyway.
Testing wasn’t even done live. Vendors were given recordings and allowed to process them offline before submitting results — creating what the auditor called the potential to “overstate their compliance with security and privacy requirements.”
What Is an AI Medical Scribe?
An AI medical scribe is an artificial intelligence tool that listens to conversations between doctors and patients and automatically generates clinical notes, including prescriptions, diagnoses, treatment plans, and referrals. It’s designed to reduce the documentation burden that consumes an estimated 35-50% of physician time. Ontario approved these systems for voluntary use in April 2025, and approximately 5,000 physicians across the province now use them.
The Government Response
Minister Stephen Crawford dismissed the findings: “The errors pointed out in the report were found during the testing phase.” He emphasised that doctors “oversee every aspect” of the AI output and that the systems are designed to save physicians time.
Health Minister Sylvia Jones did not attend the government’s news conference to respond to the report. A spokesperson said doctors “must review and approve” all documentation, and that use requires patient consent.
But here’s the problem: doctors are not required to formally attest that they reviewed the AI notes. There’s no IT control enforcing sign-off. The auditor specifically recommended implementing such an attestation requirement. The government has only agreed to “determine the feasibility” of including it in future procurements.
Green Party Leader Mike Schreiner called the findings “deeply disturbing”: “If the government’s going to utilise these tools, we need to make sure they work properly before they’re deployed.”
The Privacy Breach Precedent
This isn’t theoretical risk. In September 2024, an unapproved AI scribe caused a privacy breach that exposed hospital patient information to current and former staff. The auditor noted this happened before Ontario officially approved AI scribes, but it demonstrates the real consequences of inadequate oversight.
Ontario is asking patients to trust that their most intimate health information — their bodies, their mental health, their prescriptions — is being handled by systems that literally make things up.
Why It Matters for New Zealand
New Zealand has no equivalent audit of AI medical tools. The Health Information Standards Organisation has issued guidelines, but there’s no mandatory testing, no procurement accuracy requirements, and no auditor general scrutinising what happens when AI meets clinical reality.
What NZ should learn from Ontario:
- Test before you deploy — not vendor-submitted test results, but independent, live testing with real clinical scenarios
- Weight accuracy above domestic presence — if your AI scribe writes the wrong drug, “made in New Zealand” is cold comfort
- Require formal attestation — doctors must sign off on AI-generated notes, enforced by IT controls, not just guidelines
- Mandate third-party security audits — no approval without verified security credentials
NZ’s Health Certainty Bill, when it eventually arrives, needs to include these provisions. Ontario just showed what happens when you don’t.
🔍 THE BOTTOM LINE
Ontario deployed AI scribes that write wrong drugs, fabricate treatments, and miss mental health crises — after a procurement process that valued being local over being accurate. 5,000 doctors are using these systems right now. The auditor general personally asked her own doctor to check the transcript. If the person auditing the system doesn’t trust it with her own health data, why should anyone else?
